Vol.

9, 2024-12

Discrete Events Randomistics

Hugo Hernandez
ForsChem Research, 050030 Medellin, Colombia
[Link]@[Link]

doi: 10.13140/RG.2.2.17447.15520

Abstract
While most mathematical models representing the real World are continuous, all real systems
and processes are, in fact, discrete. Discreteness emerges, for example, by considering that
matter is comprised by discrete units and aggregates (e.g. particles, atoms, molecules, etc.).
Time, space and any other measured variable can be considered to be ultimately discrete
instead of a pure continuum simply due to our technical limitations. In addition, many
observable properties have a discrete nature (either numerical or categorical). Unfortunately,
the mathematical framework of discrete models is underdeveloped compared to that of
continuous models, probably due to the non-differentiable nature of discrete variables. As a
contribution to such mathematical framework, in this report, the mathematics of discrete
randomistic variables (simultaneously considering determinism and randomness) is presented.
Basic definitions and concepts are included as well as the analytical derivation of multiple
distribution models for different variables observed in discrete events.

Keywords
Activities, Central Limit Theorem, Change of Variable, Discrete Events, Distribution Models,
Interarrival Times, Queues, Randomistics

1. Introduction

There is a thin boundary separating the notions of discrete and continuous variables [1] or data
elements [2] in general. From a physical perspective, and due to measurement limitations
(resolution), all observable variables are discrete, and should be treated as such. However,
from a mathematical perspective, all observable variables can be represented by ideal,
continuous functions. Of course, representing a variable using a continuous function does not
necessarily imply that the variable is truly continuous.
In dynamic systems, a discrete variable experiences a “sudden” change in its value at some
instant of time. Mathematically speaking, such change represents a jump, step or first-order

Cite as: Hernandez, H. (2024). Discrete Events Randomistics. ForsChem Research Reports, 9, 2024-12, 1 -
25. doi: 10.13140/RG.2.2.17447.15520. Publication Date: 15/10/2024.
 Discrete Events Randomistics
Hugo Hernandez
ForsChem Research
[Link]@[Link]

discontinuity [3], and its time-derivative is undetermined or simply non-existent. Depending on

the scale of observation, the discrete variable can be represented by a continuous,
differentiable function. Particularly, if the time scale is much longer than the characteristic time
of the process, the continuous approximation can be considered satisfactory. For shorter time
scales, the continuous approximation might not be adequate. This situation is illustrated in
Figure 1, where a discontinuous variable can be nicely approximated by a continuous function
when longer observation times are considered.

Figure 1. Discrete process observed at different time scales. The observed variable (a sinusoidal variable)
only takes integer values, and changes in this variable are observed only at integer units of time. Top left:
Process observed during 5 arbitrary units of time. Top right: Process observed during 50 arbitrary units of
time. Bottom left: Process observed during 500 arbitrary units of time. Bottom right: Process observed
during 1000 arbitrary units of time.

Practically all real processes are discrete at some point. It was already mentioned that limited
measurement resolution introduces discreteness into the system, even at macroscopic scales.
But it is also possible to observe additional discrete macroscopic changes, like for example
turning on or off a machine, starting or finishing a certain activity, or when a variable
overcomes certain arbitrary limits (for quality or safety reasons), etc. In addition, at
microscopic scales, all processes involve discrete events taking place between individual
molecules, atoms, or even subatomic entities. We might also consider that time is ultimately
discrete and not continuous (as is assumed in computer simulations and numerical methods),
or at least, that experimental information is always collected discretely and never continuously.

15/10/2024 ForsChem Research Reports Vol. 9, 2024-12

10.13140/RG.2.2.17447.15520 [Link] / [Link]/forschem (2 / 25)
 Discrete Events Randomistics
Hugo Hernandez
ForsChem Research
[Link]@[Link]

Dealing with discontinuous, discrete events is challenging from a mathematical perspective,

particularly due to the non-differentiable nature of discrete variables. If we also consider
randomness and uncertainty, they become even more challenging. Such randomness may
emerge in the duration of each step, or due to the presence of noise in the determination of
the observed variable value. The purpose of this series of reports is discussing the
mathematical description of “randomistic” discrete events. The term randomistic is used here
to denote the integrated realm including both random and deterministic variables [4,5].
This report is devoted to explaining the fundamentals of discrete processes, introducing basic
concepts, and presenting general procedures for describing randomness in discrete event
processes. The concepts are introduced here in the most general way possible, but they can be
particularly employed to describe physical and chemical phenomena, such as molecular
collisions, molecular aggregation processes, chemical reactions, phase transitions, heat and
transfer processes, and so on.

2. Basic Concepts and Definitions

The main concepts employed in this report are defined in this section1. The starting point will be
the concept of event, which was already mentioned in the previous section.

Event: Sudden change in the value or in the behavior of a variable. The variable
considered must not necessarily be discrete. However, any change in a discrete
variable will necessarily represent an event. The variable experiencing an event,
can be associated with the state of a certain object, denoted here as a station.
Station: Individual entity or object characterized by at least one discrete state (which
can be either categorical or numerical). The most representative discrete state
considered is the activity of the station. Other discrete and continuous states
can also describe the station, like for example: activity type, queue length,
opportunity counter, success counter, and many other.
Activity: Discrete state of a station indicating whether it is active (busy) or inactive (idle).
Only two possible values of the state are considered. Thus, the activity state can
be represented by a numerical binary variable (having a value of 0 when inactive
or when active), by a Boolean variable (TRUE for active, FALSE for inactive), or
simply by any arbitrary binomial categorical variable (with only two exclusive
and exhaustive categories2).

1
Some definitions presented here may slightly differ from the definitions employed by other authors in
the field of discrete events, but the main ideas are expected to remain the same.
2
This means that at exactly one category (no more and no less) is valid at a given moment.

15/10/2024 ForsChem Research Reports Vol. 9, 2024-12

10.13140/RG.2.2.17447.15520 [Link] / [Link]/forschem (3 / 25)
 Discrete Events Randomistics
Hugo Hernandez
ForsChem Research
[Link]@[Link]

Activity Type: Discrete state of a station indicating the nature of the activity currently
performed by the station. A null category (denoted for example as “None” for
categorical variables, or as for numerical variables) might be required to
indicate that the station is inactive. If the station can only perform a single type
of activity, then this variable is not required, as it is equivalent to the activity
variable mentioned in the previous definition. Stations might also be able to
perform multiple activities simultaneously. However, stations will be assumed
capable to perform only a single activity at a time, unless otherwise stated. The
type of activity is commonly determined by the guest currently at the station.
Guest: Individual entity or object inducing an event in the station. Each guest may have
associated different randomistic variables (attributes) including the requested
activity type, the activity duration, the arrival time, the maximum wait time, etc.
These variables are considered randomistic because their values can be either
random or deterministic. Once the guest is active (is attended by a station), the
requested activity type determines the activity type of the station. A single
guest may request multiple activities in sequence as it arrives to the station.
Requested Activity Type: Discrete attribute of a guest, indicating the nature of the activity to be
performed by the station. The main difference with the activity type is that null
categories are not allowed in this case. Also, if the requested activity type is not
allowed by the station, the outcome of the activity will always be unsuccessful,
and the activity duration will be shorter (as it will be limited only to set-up and
wrap-up times).
Activity Duration: Continuous attribute of a guest, indicating the time elapsed between the
starting and ending events of the activity at the station. The activity duration
include any set-up and wrap-up times required by the activity, as well as the
corresponding processing time of the activity.
Arrival Time: Continuous attribute of a guest, indicating the precise moment when it arrives
to the station. The arrival of the guest is a discrete event, where the opportunity
counter of the station increases by . If the station is busy when the guest
arrives, the guest is placed at the end of the station queue, increasing the queue
length by . After being served, a guest might return to the station at a
different, future arrival time.
Interarrival Time: Continuous variable indicating the time between two consecutive guest
arrivals to the station.
Maximum Wait Time: Continuous attribute of a guest, indicating the maximum time elapsed
in a queue. If the time spent by a guest in the queue is greater than the
maximum wait time, the guest leaves the queue and the opportunity is lost.

15/10/2024 ForsChem Research Reports Vol. 9, 2024-12

10.13140/RG.2.2.17447.15520 [Link] / [Link]/forschem (4 / 25)
 Discrete Events Randomistics
Hugo Hernandez
ForsChem Research
[Link]@[Link]

Queue Length: Discrete (non-negative integer) state of station, indicating the number of guest
that have arrived to the station and are waiting for their turn. When a guest is
attended by the station, or when it abandons the queue, the queue length
decreases by . The minimum queue length is . A maximum queue length may
be considered. In this case, any guest arriving when the queue length is at its
maximum capacity results in a lost opportunity for the station. The next guest in
the queue (if any) will be attended immediately when the station finishes its
current activity, becoming the active guest being served. Typically, guests are
attended by the station in strict order of arrival, but different priority criteria
can be used to determine the next guest in the queue.
Opportunity Counter: Discrete (non-negative integer) state of a station indicating the number
of guests that have arrived to the station since the last reset.
Opportunity Outcome: Discrete state of a guest indicating the outcome of the activity
performed at the station. Typical binary variables ( / , FALSE/TRUE,
success/failure) can be used to describe the outcome. It is initially set as null ( ,
FALSE, failure, etc.), until a successful activity is performed by the station. The
success counter of the station is only updated after performing a successful
activity.
Success Counter: Discrete (non-negative integer) state of a station indicating the number of
guests that have been successfully attended by the station since the last reset.
Busy Time: Non-negative continuous state of a station indicating the total time spent by
the station performing activities since the last reset.
Idle Time: Non-negative continuous state of a station indicating the total inactive time of
the station since the last reset. The sum of busy time and idle time is the total
run time.
Run Time: Non-negative continuous state of a station indicating the total time elapsed
since the last reset, and measured by a suitable time-keeping system (clock).
Reset: Initialization of process variables in a system. Current time, cumulative times and
discrete counters are set to zero.
Process: Set of events taking place in a system over time.
System: Particular set of stations and guests under consideration. The system may
contain one or more stations, as well as one or more guests.
Figure 2 graphically illustrates some of the concepts introduced in this section.

15/10/2024 ForsChem Research Reports Vol. 9, 2024-12

10.13140/RG.2.2.17447.15520 [Link] / [Link]/forschem (5 / 25)
 Discrete Events Randomistics
Hugo Hernandez
ForsChem Research
[Link]@[Link]

Figure 2. Example of Discrete Events in a System (left) comprising multiple stations (boxes) and multiple
guests (circles) waiting in queues. The activity requested by each guest is denoted by its color. A Process
(right) is represented by different state variables (along the vertical axis) changing with time (horizontal
axis). Activity is a binary state variable. Activity type is a categorical state variable. Opportunity counter,
Success counter and Queue length are numerical discrete state variables.

3. Distribution Models

A system with an excess of stations, where they remain idle most of the time is inefficient. If
the number of stations in the system is decreased, the overall busy time of the remaining
stations increases, but the queue lengths and waiting times may also increase to levels where
the rate of success may decrease due to queue abandonment. An optimal configuration of the
system would be possible if the arrival time and type of activities requested by future guests
were known in advance. Of course, no one can foresee the future. However, we can build
models capable of predicting the future. These models are representations of reality, but will
never be reality itself [6], and thus, they will always involve different degrees of error.
The simplest assumption that we can make about the future events, is that they will follow a
similar pattern as past events. Thus, if we collect information about events that took place in
the past, we can obtain a distribution model of possible future events. This approach can be
used to represent different state variables or properties of the guests and stations, like for
example the number of events, the arrival time between consecutive guest, the types of
activities requested, the duration of the activities, and so on.
The distribution model obtained is denoted as randomistic, as it allows considering two
different types of models: Random and Deterministic. Deterministic models assume that the
properties of future events are known with certainty, whereas random models assume that the
properties of future events are unknown. Deterministic models can be used to compare
different system configurations for a fixed sequence of events, whereas random models can be
used to evaluate average performance and worst-case scenarios for a given system

15/10/2024 ForsChem Research Reports Vol. 9, 2024-12

10.13140/RG.2.2.17447.15520 [Link] / [Link]/forschem (6 / 25)
 Discrete Events Randomistics
Hugo Hernandez
ForsChem Research
[Link]@[Link]

configuration. Both random and deterministic models can be represented by probability

distribution functions. In the case of deterministic models, as well as discrete random models,
the probability distribution function is represented by the delta function [2]. Continuous
random models can be described by any arbitrary but valid probability distribution function.
A very important parameter of randomistic distribution models is the time horizon ( )
considered by the model. In the case of deterministic models, if the simulation time is longer
than the time horizon, the predetermined sequence of events is repeated from the beginning.
In the case of random models, the time horizon determines the size of the random sample,
with important effects on the distribution models, as will be shown here. For longer simulation
times, time is reset every time horizon and new data samples are obtained. The adequate
selection of the time horizon is critical for a successful description of the process, as illustrated
in Figure 3. This plot shows the average busy time fraction observed (since the last reset) for a
station with periodic (deterministic) activities, considering different observation times (time
horizons). The station is considered active during minutes and then it becomes inactive for
minutes. This activity sequence is repeated over days. Clearly, as the time horizon
increases, the average busy time fraction value eventually becomes stable. However, extremely
long time horizons are not necessarily better, as they may fail to describe important changes
taking place in the system or its environment. A reasonable compromise is always required.

Figure 3. Average busy time fraction as a function of observation time for a periodic sequence of events.

In this Section, randomistic distribution models will be derived for some of the most important
properties of guests and stations.

15/10/2024 ForsChem Research Reports Vol. 9, 2024-12

10.13140/RG.2.2.17447.15520 [Link] / [Link]/forschem (7 / 25)
 Discrete Events Randomistics
Hugo Hernandez
ForsChem Research
[Link]@[Link]

3.1. Distribution of Activity States

Let us denote by ( ) the activity state of a given station at time [ ], where is the
reset time (e.g. current time) and is the time horizon. Activity state values are limited to
binary variables, typically considering the discrete set { }, where represents inactivity and
represents activity.
Let us now consider that the busy time fraction of the station, for a representative past time
horizon , is . This value is assumed to represent the probability of finding the station busy at
any arbitrary moment during the future time horizon. Assuming an invariant activity probability
for the station, the activity state distribution can then be modeled by the following probability
distribution function:

( ) ( )
( ( ))
( ) ( ) { [ ]
( )
(3.1)
This distribution model is known as Bernoulli distribution, and each observation of the activity
state is denoted as a Bernoulli trial [7].

3.2. Distribution of Activity Duration

The distribution model for activity duration ( ) can be obtained from past observations of the
system. Thus, a wide range of distribution models of activity duration are possible. In this
section, we will derive only some common models of activity duration.
First, let us assume that the time horizon can be divided into time intervals of equal
duration, and that the probability of activity ( ) is the same for each time interval
independently of all other intervals. Let us also assume that inactive periods of time are always
observed between different activities. The length of each time interval ( ) would be:

(3.2)
In this case, the duration of the activity is determined by the number of active time intervals
observed before the first inactive time interval appears. Notice that an initial active time
interval is mandatory. Thus, the probability of observing an activity with duration is:

( ) ( )
(3.3)
as it is assumed that intervals are active ( ) but the ( )-th time interval (after the
first active interval) is inactive ( ). Here, the activity duration is only considered a positive
integer multiple of the time interval. Nevertheless, in the case of , there is no activity, and
we simply obtain the probability of an inactive interval ( ) .

15/10/2024 ForsChem Research Reports Vol. 9, 2024-12

10.13140/RG.2.2.17447.15520 [Link] / [Link]/forschem (8 / 25)
 Discrete Events Randomistics
Hugo Hernandez
ForsChem Research
[Link]@[Link]

The distribution model given by Eq. (3.3) is known as geometric distribution, and is a particular
case of the more general negative binomial distribution [8].
Notice that Eq. (3.3) can be alternatively expressed as:

( )
( ) ( ) ( )
(3.4)
This means that the activity duration distribution model can be approximated by a continuous
exponential function. As decreases ( ), this exponential function can be associated
with a probability density function ( ) as follows:

( )
( )
( ) ( )
(3.5)
Since (because ), the natural logarithm can be approximated as a series
expansion as follows3:

( )
∑

(3.6)
Now, truncating the expansion after the first term and replacing in Eq. (3.5), the following
approximation is obtained:

( )
( ) ( )
(3.7)
which corresponds to the exponential distribution [9] where ( ) is the average rate of
occurrence of activities. For this scenario, the mean activity duration 〈 〉 is the reciprocal of
4

this rate of occurrence of activities:

〈 〉
( )
(3.8)
The exponential distribution model is commonly employed to represent the elapsed time
between consecutive events when they occur continuously and independently at a constant
average rate [10]. However, in the case of event duration, this distribution model does not take

3
Using a series expansion valid for . See: [Link] accessed on
October 1, 2024.
4
The mean activity duration 〈 〉 is determined as the expected value of the activity duration considering
the corresponding distribution model: 〈 〉 ∫ ( ) .

15/10/2024 ForsChem Research Reports Vol. 9, 2024-12

10.13140/RG.2.2.17447.15520 [Link] / [Link]/forschem (9 / 25)
 Discrete Events Randomistics
Hugo Hernandez
ForsChem Research
[Link]@[Link]

into account any minimum duration due to set-up or wrap-up operations associated to the
activity.
Considering a minimum activity duration time , the probability distribution function
then becomes:
( )
( ) {

(3.9)
For , the probability function can be represented as the exponential:

( )( ( ))
( ) ( )
(3.10)
and this can be approximated for by the following continuous probability density
function:
( )( )
( ) ( )
(3.11)
corresponding to an exponential distribution displaced a time .
The mean activity duration in this case is:

〈 〉
( )
(3.12)
A limiting, maximum activity duration ( ) can also be considered (for example, limiting the
activity duration to the time horizon, ), simply by correcting the probability density
model given in Eq. (3.11). The correction factor guarantees the consistency of the distribution
model, by setting as the total probability of the sample space. The probability density model
obtained is then:

( )( )

( ) ( )
( )( )

(3.13)
with mean activity duration:
( )( )
( )
〈 〉
( )( )

(3.14)

15/10/2024 ForsChem Research Reports Vol. 9, 2024-12

10.13140/RG.2.2.17447.15520 [Link] / [Link]/forschem (10 / 25)
 Discrete Events Randomistics
Hugo Hernandez
ForsChem Research
[Link]@[Link]

3.3. Distribution of the Number of Activities

Considering the time horizon , the mean number of activities performed 〈 〉 would be:

〈 〉
〈 〉
(3.15)
However, for a sample of activities performed during a fixed time interval , the observed
number of activities will be:

̅̅̅
(3.16)
where ̅̅̅ is the average duration of the activities in the sample.
If the time horizon is much larger than the mean activity duration ( 〈 〉), from the central
limit theorem [11], the average activity duration can be represented by a normal distribution
with a mean value of 〈 〉 and variance ( ) .

The variance of the activity duration for the most general model considered in the previous
section (Eq. 3.13) would be:

( )( )

( ) ∫ ( ) 〈 〉
( )( )

( )( )
( )( )
( ) ( ) ( )
( )( )

(3.17)
So, the sample average duration of activities (assuming 〈 〉) can be approximated in
terms of the standard normal random variable ( ) [5] as follows:

( )
̅̅̅ 〈 〉 √
〈 〉
(3.18)
Now, the reciprocal of the average activity duration will be [12]:

( )
∑ [ (〈 〉 √ ) ]
̅̅̅ 〈 〉
( )
〈 〉 √
〈 〉
(3.19)
Truncating the series expansion after the linear term we obtain:

15/10/2024 ForsChem Research Reports Vol. 9, 2024-12

10.13140/RG.2.2.17447.15520 [Link] / [Link]/forschem (11 / 25)
 Discrete Events Randomistics
Hugo Hernandez
ForsChem Research
[Link]@[Link]

( ) ( )
√ √
̅̅̅ 〈 〉 〈 〉 〈 〉 〈 〉 〈 〉 〈 〉
(3.20)
The sign of the second term can be changed due to the symmetry about of the standard
normal random variable.
Replacing Eq. (3.20) and (3.15) in Eq. (3.16) we obtain:

〈 〉 ( )
⟦〈 〉 √ ⟧
〈 〉

(3.21)
where ⟦ ⟧ represents the closest integer rounding operator.
Now, for the case where and , Eq. (3.21) simplifies into:

⟦〈 〉 √〈 〉 ⟧
(3.22)

3.4. Distribution of Activity Types

The activity type is a categorical state. Those categories should be exclusive (an activity cannot
belong to two different categories at the same time) and exhaustive (an activity must belong
to at least one category). The most direct way to model the probability distribution of activity
types ( ) is assuming that the probability of occurrence of each category is exactly the
relative frequency observed for each activity type in the past.
If such information is not available, the simplest approach is assuming a uniform distribution of
probabilities for each category [13]:

( )

(3.23)
where is the total number of activity types.
Of course, this is not necessarily the case in many real situations, and thus, the use of past
information is always strongly advised.
Activity types can also be associated to the nature of the guest ( ). Different guest categories
(which must also be exclusive and exhaustive) may have different conditional probability
distributions of activity types ( ( ) ). In those cases, the probability distribution of guest
types ( ( )) is needed, and it can be simply associated to the current fraction of guests of

15/10/2024 ForsChem Research Reports Vol. 9, 2024-12

10.13140/RG.2.2.17447.15520 [Link] / [Link]/forschem (12 / 25)
 Discrete Events Randomistics
Hugo Hernandez
ForsChem Research
[Link]@[Link]

each type in the whole guest population. Thus, the total probability of each activity type is
determined as follows (from the law of total probability [14]):

( ) ∑ ( ) ( )

(3.24)
where is the total number of guest types. Here, we have assumed that guest categories are
represented by ordinal numbers, arbitrarily assigned to each guest type.
If the conditional probability ( ) is assumed uniform for all guest types, then Eq. (3.24)
results again in Eq. (3.23).
Now, if the duration of the different activity types are different, we have:

( ) ∑ ( ) ( )

(3.25)
where ( ) represents the probability of activity duration conditioned to the activity type. If
this conditional probability distribution is unknown, it can be approximated by the following
continuous approximation (from Eq. 3.13):
( )
( )( ( ))
( )
( ) ( ) ( )
( )
( )( ( ) ( ))

( ) ( )
(3.26)
with mean duration of activity :
( )
( )( ( ) ( ))
( ( ) ( ))
〈 〉
( ) ( )
( )( ( ) ( ))

(3.27)
Notice that the busy time fraction is corrected by the probability of the activity type, since only
the fraction of time spent on the corresponding activity type is considered. Also, the minimum
( ( ) ) and maximum ( ( ) ) activity duration can be different for each activity type.

Then, the overall mean activity duration will be:

〈 〉 ∑ ( )∫ ( ) ∑ ( )〈 〉

(3.28)

15/10/2024 ForsChem Research Reports Vol. 9, 2024-12

10.13140/RG.2.2.17447.15520 [Link] / [Link]/forschem (13 / 25)
 Discrete Events Randomistics
Hugo Hernandez
ForsChem Research
[Link]@[Link]

and its variance:

( ) ∑ ( ) ( )

(3.29)
On the other hand, the number of activities of each type observed during a time horizon will
be:

( ) ⟦〈 〉 √〈 〉 ⟧
(3.30)
where

〈 〉
〈 〉
(3.31)
Then, it can be shown that the overall number of activities is:

∑ ( ) ( ) ⟦〈 〉 〈 〉 ⟧

(3.32)
These results are consistent with the models derived in Sections 3.2 and 3.3.

3.5. Distribution of Activity Outcomes

The outcome of an activity ( ) is another binary variable, which can be represented by the
values and (FALSE / TRUE, failure / success, reject / approve, no-pass/pass, etc.). The activity
outcome can be modeled in the same way as the activity state; that is, using a Bernoulli
distribution. In this case, the probability distribution of the activity outcome depends on the
activity type, and can be mathematically expressed as follows:
( )
( ) ( )( ( )) {
( )
(3.33)
where ( ) represents the success rate of activities of type .
Then, the overall activity outcome distribution will be:

〈 〉
( ) ∑ ( ) ( ) {
〈 〉

(3.34)
where 〈 〉 ∑ ( ) ( ) represents the overall success rate.

15/10/2024 ForsChem Research Reports Vol. 9, 2024-12

10.13140/RG.2.2.17447.15520 [Link] / [Link]/forschem (14 / 25)
 Discrete Events Randomistics
Hugo Hernandez
ForsChem Research
[Link]@[Link]

3.6. Distribution of Time between Consecutive Events

As we have seen in Section 3.3, considering an interval of time identical to the time horizon ,
the number of activities performed by a station is , which is an integer variable
approximately normally distributed (Eq. 3.22). If the system conditions remain identical over
the whole interval , the beginning of those activities can be assumed to be uniformly
distributed along the interval. This assumption guarantees that the rate of activity execution
remains relatively constant during the time interval. Of course, since the events are discrete,
instantaneous rates will never be constant. However, average rates over longer observation
times will eventually become stable for uniformly distributed activities.
Thus, for the time interval [ ], the distribution of initial activity times ( ) can be
assumed to follow the uniform probability density model:

( )

(3.35)
where the cumulative probability distribution is:

( ) ( ) ∫

(3.36)
Now, considering a random sample of activities, and by assuming that the first activity
began exactly at time , then we obtain uniform random time values in the interval.
Then, the next activity will begin at the minimum value in the random sample (
( )), having the following distribution [15]:

( ) ( ( )) ( )
(3.37)
The corresponding probability density function for the next activity time can be obtained from
the time derivative of Eq. (3.37):

( ) ( )
( ) ( )

(3.38)
where Eq. (3.38) can be equivalently expressed as follows:
( ) ( ) ( )
( )

(3.39)
And the distribution of times between consecutive or sequential activities ( ), where
, would be:

15/10/2024 ForsChem Research Reports Vol. 9, 2024-12

10.13140/RG.2.2.17447.15520 [Link] / [Link]/forschem (15 / 25)
 Discrete Events Randomistics
Hugo Hernandez
ForsChem Research
[Link]@[Link]

( ) ( ) ( )
( )

(3.40)

Notice that for , we find that , ( ) and is large, and therefore,

Eq. (3.40) can be approximated by the following exponential distribution model:

( )
( ) ( )
(3.41)
Also notice that no minimum time between activities was considered in this approach. So, if the
minimum activity duration is considered, the resulting sequential time distribution model
(for uniform distributions of initial activity times and infinite time horizons) would be:

( )( )
( ) ( )
(3.42)
In terms of standard random variables,

( )

(3.43)
where is the type II standard exponential random variable [16].

A similar procedure can be performed to obtain the distribution of time between consecutive
activities considering initial time distributions along the time horizon other than the uniform
distribution. In addition, this same approach can be used to model the distribution of time
between consecutive events in general, regardless of the type of event, as long as the
distribution of event times over a certain time horizon is known. For example, it is possible to
describe the time between two consecutive opportunities, that is, the arrival time difference
between two consecutive guests at a given station.

3.7. Distribution of Time between Consecutive Successful Events

Now that we have obtained a model for the distribution of time between consecutive events or
activities, let us consider the outcome of such event or activity.
Previously we have denoted the overall success rate in a station by 〈 〉, representing the
probability of obtaining a successful outcome in any activity performed in such station. For
〈 〉 , all activities performed by the station will be successful, and therefore the distribution
of time between successful events is identical to the distribution of time between consecutive
activities. On the contrary, for 〈 〉 , not all activities will be successful.

15/10/2024 ForsChem Research Reports Vol. 9, 2024-12

10.13140/RG.2.2.17447.15520 [Link] / [Link]/forschem (16 / 25)
 Discrete Events Randomistics
Hugo Hernandez
ForsChem Research
[Link]@[Link]

The number of unsuccessful activities ( ) performed until a successful activity occurs can be
described by a geometric distribution [8], as follows:
〈 〉( 〈 〉) 〈 〉 ( 〈 〉)
( )
(3.44)
which can be approximated for 〈 〉 by the discrete exponential distribution:
〈 〉 〈 〉
( )
(3.45)
In terms of standard random variables we have:

⟦ ⟧
〈 〉
(3.46)
where is the type II standard exponential random variable [16] and ⟦ ⟧ is the closest-
integer rounding operator.
If unsuccessful activities are performed between two successful activities, then the time
between two consecutive successful activities ( ) will be the sum of times between
consecutive activities ( ) (considering the exponential approximation presented in Eq. 3.43):

∑ ∑ ( ( ) ) ( ) ( ) ∑

(3.47)
Now, the last sum term in Eq. (3.47) represents the Erlang distribution model [17]. Thus, the
probability density function of becomes:

( )
( )
( ) ( )
( )
( )
( ) ( )
( )
(3.48)
In terms of standard random variables:

( )( ( ) ( ))

(3.49)

where ( ) represents the type II standard Erlang random variable with degrees
of freedom (also known as shape parameter), with (
, , and
( )) ( ( ))

probability density function:

15/10/2024 ForsChem Research Reports Vol. 9, 2024-12

10.13140/RG.2.2.17447.15520 [Link] / [Link]/forschem (17 / 25)
 Discrete Events Randomistics
Hugo Hernandez
ForsChem Research
[Link]@[Link]

( )
(( ) ( ))
( )

( ( )) ( )
(3.50)
As increases, the time between consecutive successful events approximates the following
normal distribution (from the central limit theorem [11]):
√
( )( ) ( )

(3.51)
Notice that, from Eq. (3.46), increases when 〈 〉 decreases. That is, when the rate of success
decreases. Also notice that is a random variable and therefore the distribution of the time
between successful events is a function of two independent random variables ( and ):

√ ⟦ ⟧
〈 〉
( ⟦ ⟧) ( ) 〈 〉
〈 〉

( )
(3.52)
If the success rate decreases even further, and notice that it was previously assumed that
〈 〉 , the time between successful events approaches the following discrete-exponential
distribution:

( ) ( )⟦ ⟧ 〈 〉
〈 〉
(3.53)
Interestingly, the time between consecutive successful events does not necessarily follow the
same distribution of the time between consecutive events.

3.8. Cumulative Number of Guest Arrivals

Assuming a uniform distribution of times for the occurrence of events, the cumulative number
of guest5 arrivals ( ( ) ) taking place in the time interval [ ] would increase almost linearly
over time ( ) during the time horizon ( ), as can be concluded from Eq. (3.36):
( )
( ) ( ) ( ) ( )
(3.54)
In general, for any other type of time distribution ( ), we may obtain the following relation:

5
The notation is used for guests instead of , which was used for activities.

15/10/2024 ForsChem Research Reports Vol. 9, 2024-12

10.13140/RG.2.2.17447.15520 [Link] / [Link]/forschem (18 / 25)
 Discrete Events Randomistics
Hugo Hernandez
ForsChem Research
[Link]@[Link]

( ) ( ) ( ) ∫ ( )

(3.55)

Or alternatively, if the distribution of times between consecutive events ( ( )) is known, the

cumulative number of events would be (from Eq. 3.37 and 3.38, applied to guests instead of
activities):

( )
( ) ( )( ( ∫ ( ) ) )

(3.56)
For example, assuming an exponential distribution of time between consecutive guest arrivals,
the cumulative number of guest arrivals becomes:

( ) ( )
( ) ( )
( ) ( )( ( ∫ ( ) ) ) ( )( )

(3.57)
Furthermore, for , the cumulative number of guest arrivals can be approximated by
the following linear model:
( )
( ) ( )
(3.58)
which is consistent with the result obtained in Eq. (3.54).

3.8. Distribution of Queue Lengths and Event Rates

The length of the queue at a station ( ) depends on the number of guests arriving to the
station ( ) as well as on the number of guests being served ( ) or withdrawing before being
attended ( ). In mathematical terms we have:

( ) ( ) ( ) ( )
(3.59)
where the terms , , and are cumulative numbers, and the operator indicates a
forward difference over a certain period.
For example, if we consider a time difference , the change in the cumulative number of
guests arriving to the station will be:

15/10/2024 ForsChem Research Reports Vol. 9, 2024-12

10.13140/RG.2.2.17447.15520 [Link] / [Link]/forschem (19 / 25)
 Discrete Events Randomistics
Hugo Hernandez
ForsChem Research
[Link]@[Link]

( ) ( ) ( )
(3.60)
Now, if we assume a uniform distribution of arrival times, then from Eq. (3.54):
( ) ( ) ( )
( ) ( ) ( ) ̅
(3.61)
( )
where ̅ represents the average rate of guest arrivals over the time interval . This
concept might be further expanded by considering the frequency or instantaneous rate of
guest arrivals as follows (for ):
( )
( )

(3.62)
However, since ( ) is a discrete, discontinuous variable, its derivative is formally inexistent.
( ) can be considered only as an approximate continuous model, but not as a true
instantaneous rate of guest arrivals. To better understand this model, let us consider that:

( ) ( ) ( ) ( )
(3.63)
This means that:

( )
( )
(3.64)
If we assume, for example, an exponential distribution of guests inter-arrival times (with a
minimum time ), then the frequency of guest arrivals will be a random variable, independent
of time as follows:

( )
( ) ( )
( )
(3.65)
where is the minimum inter-arrival time between consecutive guests, and ( ) is the
number of guests arriving over the time horizon considered.
A more precise result is obtained considering the change of variable theorem [18], which is also
valid for . In this case, the probability density function of the instantaneous rate of
guest arrivals is:

15/10/2024 ForsChem Research Reports Vol. 9, 2024-12

10.13140/RG.2.2.17447.15520 [Link] / [Link]/forschem (20 / 25)
 Discrete Events Randomistics
Hugo Hernandez
ForsChem Research
[Link]@[Link]

( )
( )
( )
( )

(3.66)
For , Eq. (3.66) corresponds to an inverted chi-squared distribution (a particular case of
the inverted gamma distribution) [19] with degrees of freedom.
Now notice that the average rate of guest arrivals is not the same as the average of the
instantaneous rate of guest arrivals:
( ) ( )
̅ ( ) ( )
( )

(3.67)
where represents the incomplete gamma function [20].
The difference between both terms is that ̅ actually represents the reciprocal of the average
time between consecutive guest arrivals:
( )
̅
( ) ( )
( )
(3.68)
( )
For this reason, when then ( ) whereas ̅ . This situation corresponds to
a frequency paradox [21]. On the other hand, when , we obtain similar results in both
( )

cases: ( ) and ̅ .

So far, we have assumed a constant average rate of guest arrival. However, in real processes,
the event rate may change depending on external or internal conditions. Thus, the rate of
arrivals can be expressed as a function of those internal or external factors. For example, in the
case of guest arrivals, we may assume that the rate of arrival depends on the number of
potential guests close to the station. While such relation can be assumed linear, non-linear
relations may also occur. For the linear relation we have:

( ) ( ) ( )
(3.69)
where ( ) is a rate constant, and ( ) is the number of potential guests close to the station.

Similar considerations can be done for the number of guests being served and the number of
guests abandoning the queue. However, it is important to notice that those events may only
take place when there is at least one guest at the station queue.
The main advantage of considering a frequency or instantaneous rate of events is that Eq.
(3.59) can be transformed into a differential equation as follows:

15/10/2024 ForsChem Research Reports Vol. 9, 2024-12

10.13140/RG.2.2.17447.15520 [Link] / [Link]/forschem (21 / 25)
 Discrete Events Randomistics
Hugo Hernandez
ForsChem Research
[Link]@[Link]

( )
( ) ( ) ( )

(3.70)
Thus, the probability distribution model for the queue length can be obtained by using the
change of variable theorem [18], as follows:

( ) ( )
| | | ( ) ( ) ( )|

(3.71)

For example, if the term is a positive constant, then the distribution of queue lengths would
be simply uniform. Negative constant values of do not result in uniform distributions
because negative queue lengths are unfeasible. For example, if ( ) ( ) ( ),
then a Poisson distribution [22] of queue lengths would be obtained, also equivalent to a
Binomial distribution [23] with a low probability of success and a large number of trials.

On the other hand, the probability density of queue lengths would be maximal when ,a
situation representing a steady-state condition.
Also notice that the queue length can be obtained by integrating Eq. (3.70) as follows:

( ) ⟦∫ ( ( ) ( ) ( )) ⟧

(3.72)
Considering individual rates as random variables, then the integral in Eq. (3.72) considering the
whole time horizon will also represent a random variable with the following properties [24]:

〈 〉 ( ) 〈 〉
(3.73)
〈 〉
( ) 〈 〉
( )

(3.74)
where 〈 〉 is the mean time between consecutive events.

Now, the case where 〈 〉 can be approximated by a discrete normal distribution

(considering the central limit theorem [11]) with a mean value 〈 〉 ( ) and variance
〈 〉
( ) . The discrete normal distribution can also be represented by a Poisson
( )

distribution or by a uniform binomial distribution when 〈 〉 .

15/10/2024 ForsChem Research Reports Vol. 9, 2024-12

10.13140/RG.2.2.17447.15520 [Link] / [Link]/forschem (22 / 25)
 Discrete Events Randomistics
Hugo Hernandez
ForsChem Research
[Link]@[Link]

4. Conclusion
Many different probability distribution models can be obtained to describe the future,
uncertain behavior of discrete processes. Individual predictions of the future behavior of a
discrete process can be obtained by (Monte Carlo) simulation [25]. Understanding the
uncertainty of the model requires performing a large number of independent simulations.
Nevertheless, it would always be desirable to describe random discrete events analytically. The
purpose of randomistics is providing an analytical framework to describe random variables,
which will also be valid for deterministic variables. In that sense, the analytical formulation of
different distribution models associated to discrete event systems was derived and presented
in this report. These analytical, randomistic models can be used to analyze discrete processes,
as an alternative, or a complement, to simulation.

Acknowledgment and Disclaimer

This report provides data, information and conclusions obtained by the author(s) as a result of original
scientific research, based on the best scientific knowledge available to the author(s). The main purpose
of this publication is the open sharing of scientific knowledge. Any mistake, omission, error or inaccuracy
published, if any, is completely unintentional.

This research did not receive any specific grant from funding agencies in the public, commercial, or not-
for-profit sectors.

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC
4.0). Anyone is free to share (copy and redistribute the material in any medium or format) or adapt
(remix, transform, and build upon the material) this work under the following terms:
 Attribution: Appropriate credit must be given, providing a link to the license, and indicating if
changes were made. This can be done in any reasonable manner, but not in any way that
suggests endorsement by the licensor.
 NonCommercial: This material may not be used for commercial purposes.

References

[1] Hernandez, H. (2020). On the Discreteness of Measured Variables and the Continuous
Approximation. ForsChem Research Reports, 5, 2020-20, 1-18. doi: 10.13140/RG.2.2.27740.00646.
[2] Hernandez, H. (2024). Randomistic Data Elements. ForsChem Research Reports, 9, 2024-11, 1 - 25.
doi: 10.13140/RG.2.2.22772.36485.
[3] Gear, C. W., & Osterby, O. (1984). Solving ordinary differential equations with discontinuities. ACM
Transactions on Mathematical Software (TOMS), 10 (1), 23-44. doi: 10.1145/356068.356071.
[4] Hernandez, H. (2018). The Realm of Randomistic Variables. ForsChem Research Reports, 3, 2018-
10, 1-19. doi: 10.13140/RG.2.2.29034.16326.

15/10/2024 ForsChem Research Reports Vol. 9, 2024-12

10.13140/RG.2.2.17447.15520 [Link] / [Link]/forschem (23 / 25)
 Discrete Events Randomistics
Hugo Hernandez
ForsChem Research
[Link]@[Link]

[5] Hernandez, H. (2022). Standard Deterministic, Standard Random, and Randomistic Variables.
ForsChem Research Reports, 7, 2022-06, 1 - 18. doi: 10.13140/RG.2.2.36316.87688.
[6] Hernandez, H. (2023). Question Everything: Models vs. Reality. ForsChem Research Reports, 8,
2023-07, 1 - 11. doi: 10.13140/RG.2.2.35224.67845.
[7] Forbes, C., Evans, M., Hastings, N., & Peacock, B. (2011). Statistical Distributions. 4th Ed. John Wiley
& Sons, Inc., Hoboken, NJ. Chapter 7: Bernoulli Distribution, pp. 53-54. ISBN: 978-0-470-62724-2.
[8] Forbes, C., Evans, M., Hastings, N., & Peacock, B. (2011). Statistical Distributions. 4th Ed. John Wiley
& Sons, Inc., Hoboken, NJ. Chapter 23: Geometric Distribution, pp. 114-116. ISBN: 978-0-470-62724-
2.
[9] Forbes, C., Evans, M., Hastings, N., & Peacock, B. (2011). Statistical Distributions. 4th Ed. John Wiley
& Sons, Inc., Hoboken, NJ. Chapter 17: Exponential Distribution, pp. 77-81. ISBN: 978-0-470-62724-
2.
[10] Tomy, L., Jose, M., & Veena, G. (2020). A review on recent generalizations of exponential
distribution. Biometrics & Biostatistics International Journal, 9 (4), 152-156. doi:
10.15406/bbij.2020.09.00313.
[11] Hernandez, H. (2019). Sums and Averages of Large Samples Using Standard Transformations: The
Central Limit Theorem and the Law of Large Numbers. ForsChem Research Reports, 4, 2019-01, 1-
14. doi: 10.13140/RG.2.2.32429.33767.
[12] Hernandez, H. (2016). Modelling the effect of fluctuation in nonlinear systems using variance
algebra - Application to light scattering of ideal gases, ForsChem Research Reports, 1, 2016-1, 1-19.
doi: 10.13140/RG.2.2.36501.52969.
[13] Forbes, C., Evans, M., Hastings, N., & Peacock, B. (2011). Statistical Distributions. 4th Ed. John Wiley
& Sons, Inc., Hoboken, NJ. Chapter 41: Rectangular (Uniform) Discrete Distribution, pp. 180-182.
ISBN: 978-0-470-62724-2.
[14] Hsu, H. (2020). Schaum’s Outline of Probability, Random Variables, and Random Processes. 4th
Edition. McGraw-Hill, New York. Section 1.7. Total Probability. ISBN: 9781260453812.
[15] Bobotas, P., & Koutras, M. V. (2019). Distributions of the minimum and the maximum of a random
number of random variables. Statistics & Probability Letters, 146, 57-64. doi:
10.1016/[Link].2018.10.023.
[16] Hernandez, H. (2018). Expected Value, Variance and Covariance of Natural Powers of
Representative Standard Random Variables. ForsChem Research Reports, 3, 2018-08, 1-19. doi:
10.13140/RG.2.2.15187.07205.
[17] Forbes, C., Evans, M., Hastings, N., & Peacock, B. (2011). Statistical Distributions. 4th Ed. John Wiley
& Sons, Inc., Hoboken, NJ. Chapter 15: Erlang Distribution, pp. 84-85. ISBN: 978-0-470-62724-2.
[18] Hernandez, H. (2017). Multivariate Probability Theory: Determination of Probability Density
Functions. ForsChem Research Reports, 2, 2017-13, 1-13. doi: 10.13140/RG.2.2.28214.60481.
[19] Bernardo, J. M., & Smith, A. F. M. (2000). Bayesian Theory. John Wiley & Sons, New York. Section
3.2.2. Some particular univariate distributions. p. 119. ISBN: 0-471-49464-X.
[20] Weisstein, E. W. (2024). Incomplete Gamma Function. From MathWorld--A Wolfram Web
Resource. [Link] Accessed on
October 9, 2024.
[21] Hernandez, H. (2017). Multicomponent Molecular Collision Kinetics: Collision Rate and the Collision
Frequency Paradox. ForsChem Research Reports, 2, 2017-8, 1-21. doi: 10.13140/RG.2.2.32983.27048.
[22] Forbes, C., Evans, M., Hastings, N., & Peacock, B. (2011). Statistical Distributions. 4th Ed. John Wiley
& Sons, Inc., Hoboken, NJ. Chapter 35: Poisson Distribution, pp. 152-156. ISBN: 978-0-470-62724-2.

15/10/2024 ForsChem Research Reports Vol. 9, 2024-12

10.13140/RG.2.2.17447.15520 [Link] / [Link]/forschem (24 / 25)
 Discrete Events Randomistics
Hugo Hernandez
ForsChem Research
[Link]@[Link]

[23] Forbes, C., Evans, M., Hastings, N., & Peacock, B. (2011). Statistical Distributions. 4th Ed. John Wiley
& Sons, Inc., Hoboken, NJ. Chapter 9: Binomial Distribution, pp. 62-65. ISBN: 978-0-470-62724-2.
[24] Hernandez, H. (2018). Integrating Functions of Random Variables. ForsChem Research Reports, 3,
2018-07, 1-26. doi: 10.13140/RG.2.2.23660.87680.
[25] Metropolis, N., & Ulam, S. (1949). The Monte Carlo method. Journal of the American Statistical
Association, 44 (247), 335-341. doi: 10.1080/01621459.1949.10483310.

15/10/2024 ForsChem Research Reports Vol. 9, 2024-12

10.13140/RG.2.2.17447.15520 [Link] / [Link]/forschem (25 / 25)