Download as pdf or txt
Download as pdf or txt
You are on page 1of 16

Improving the Duckworth Lewis Stern Method for Better Outcomes in

Rain-Affected Cricket Matches

The Duckworth Lewis Method

The DLS (Duckworth Lewis Stern) method (previously the D/L method before being adjusted
by Steven E. Stern) is a mathematical formulation designed to calculate the target score for the
team batting second in a rain-affected limited overs cricket match. The method was first used
internationally on 1st January 1997 and was formally adopted by the ICC (International Cricket
Council) as the standard method of calculating target scores in rain-affected matches since
1999. However, the method has brought controversies into the world of cricket and is not
without flaws. Therefore, it is necessary to ascertain whether the DLS method should really be
used worldwide, and to explore the possibility of superior alternatives that could be used to
eliminate such controversial events.

The DLS method was first introduced use as a reaction to the 1992 World Cup Semi-Final,
between England and South Africa, in which South Africa’s chances of qualifying for the final
was eliminated by the ‘most productive overs’ method, which set the altered target score using
the highest scoring overs from the first innings. This method essentially punished the chasing
team for bowling economical overs, as they were not taken into consideration. The method
took almost no further factors into account and was regarded by the cricket community as ‘non-
scientific.’ As a result, in this game, South Africa’s target of 22 runs from 13 balls was reduced
to 21 runs from 1 ball, which meant South Africa went from a position of possibly victory into
a position of certain loss. This event highlighted the need for proper scientific modelling to
replace intuitive approaches with the resetting targets in rain-interrupted matches (Stern, 2016).

The original D/L (Duckworth Lewis) method provided a more ‘scientific’ approach than the
aforementioned ‘most productive overs’ method, using the idea of the percentage of resources
utilised, using the numbers of overs and wickets remaining to determine how many resources
a team has already used. A team has no more resources available if they have played the
maximum overs or have lost all of their wickets. If 𝑅! and 𝑅" are the percentage of resources

Sai Mehta (L6R1) 1


utilised for the first and second innings respectively and 𝑆! is the first innings score, the target
score is, 𝑆! 𝑅" /𝑅! .

Determining the percentage of resources utilised takes advantage of a resource table, which
provides this value with the information of the percentage of overs left and the number of
wickets remaining.

Figure 1: The D/L method of adjusting target scores in interrupted one-day cricket
matches (Duckworth & Lewis, 2004)

The table in Figure 1 which shows how the adjusted target score is determined during a rain-
affected cricket match is created using the following formula:

𝑍(𝑢, 𝑤) = 𝑍# 𝐹$ {1 − 𝑒 %&'/)! }

In this formula, 𝑍(𝑢, 𝑤) represents the average runs scored in the remaining overs available,
where 𝑢 is the number of overs available and 𝑤 is the number of wickets lost. 𝑍# and 𝑏 are
positive constants, and 𝐹$ is a positive decreasing step function where 𝐹# = 1. From this we

Sai Mehta (L6R1) 2


can see that the system operates on the basis of exponential decay, with the theory that each
batsman has a better run-scoring ability than the next batsman to come.

The Duckworth Lewis Stern Method

In 2003, the ICC adopted the DLS method. This method, adjusted from the D/L by Steven E.
Stern, adopts a similar approach of exponential decay, however, as part of this ‘dampening
exponent’, 𝑛# , is introduced. This was done in response to higher scoring rates in the modern
game. Stern found that whilst the D/L method fitted the median data well, it was not as good
for higher scoring games. He also found that for higher scoring games, the proportion of runs
scored at the start of the innings was greater.

The adjusted DLS formula is as follows:

"!
𝑍*+, (𝑢, 𝑤, 𝜆) = 𝑍# 𝐹$ 𝜆-! .! {1 − 𝑒 %&'/)! / }

where the ‘match factor’, 𝜆, is defined as 𝑍(𝑀, 0, 𝜆) = 𝑆! where the game is an 𝑀-over match
and 𝑛$ = 𝑛# 𝐹$ . (Stern, 2016)

Jayadevan’s Criticism

In V. Jayadevan’s criticism of the method, he proposed a regression-based approach. In his


view, 𝑍(𝑢, 𝑤) would be better calculated using a cubic function rather than an exponential one.
In Figure 2, below, Curve 2 is a representation of his regression-based method. Although the
D/L method provides a family of curves rather than one, Jayadevan used Curve 3 as an
associated curve of the D/L method.

The formula he suggested turned out to be:

𝑅 = 1.6631192𝑂 − 0.009254𝑂" + 0.0000261𝑂0

where 𝑅 represents the percentage of runs scored and 𝑂 represents the percentage of overs
bowled. (Jayadevan, 2002)

Sai Mehta (L6R1) 3


Figure 2: Comparison between D/L method and Jayadevan’s proposed method.
(Jayadevan, 2002)

Looking at Figure 2, it can be seen that the gradient of Curve 3 is constantly increasing, due to
the fact that the derivative of an exponential curve is itself exponential as well, suggesting that
the batting side is constantly accelerating. In Curve 2, this acceleration is not always positive,
and the team only begins to accelerate after roughly 60-70% of the overs have been bowled.
The reality is closer to Curve 2 than Curve 3, as the batting side generally splits up its innings
into phases rather than accelerating throughout the whole innings, mainly due to the field
restrictions in place at the start of the innings, allowing run-scoring to occur at a faster rate.
Although the exponent of an associated DLS curve is less than that of Curve 3 because of the
dampening, an associated DLS curve would follow the same pattern. As a result, the current
method in place for rain-affected matches is generally in favour of the batting side, as it
assumes the batting side will end up scoring at an even higher rate than it normally would at
the end of the innings. Not only does this give the team who bats second an advantage, it also
means that more captains will elect to field first with rain forecasted for the second innings, as
it means they will have to chase a lower score, also keeping in mind that higher levels of
precipitation causes the ball to become more slippery, thereby making it more difficult for

Sai Mehta (L6R1) 4


bowlers to execute their plans, especially in the latter phase of the batting innings where
bowlers combat aggressive batting with ‘death’ bowling – yorkers will become harder to
execute and so batting is easier in such conditions.

However, Jayadevan’s approach also assumes that the acceleration will occur at a certain point
in the innings. It only makes sense for the innings to accelerate if the batting side intends to do
so. This acceleration can also only happen if the team is able to increase its scoring rate when
it tries to, but this is also linked with the intent factor as already mentioned. For example, a
batsman who has been at the crease for a reasonable amount of time is more likely to be more
confident and so to have more intent but is also more likely to be able to score more boundaries
due to the fact that the batsman has gained more experience at the crease.

The aim of the project is to evaluate the DLS method and seek an alternative solution to the
problem.

A Proposed Solution

We can define a point in a cricket match using three general parameters – the number of overs
bowled, the number of wickets lost, and the total number of runs scored. Let the variables 𝑂,
𝑊 and 𝑅 represent these respectively. Although these are well-defined parameters, a potential
problem is that they all interact with one another, making it difficult to calculate the total
number of runs scored at the end of an innings as a function of overs and wickets. The run rate
is affected by all three factors – as 𝑂 increases, a team will attempt to accelerate and increase
its run rate, as 𝑊 increases, a team will reduce its run rate to reduce risk, and as 𝑅 increases,
team confidence increases, and so the run rate increases. Therefore, the run rate can be
expressed as some function of the three parameters:

𝑑𝑅
= 𝑢(𝑊, 𝑅, 𝑂)
𝑑𝑂

The ‘wicket rate’, which will be defined as the rate of wicket loss, is also affected by all three
factors – as 𝑂 increases, the value of a wicket decreases so teams will be happier to give wickets
away, as 𝑊 increases, a team will try to play more conservatively and the wicket rate should
decrease and as 𝑅 increases, team confidence increases, and so more risks will be taken which

Sai Mehta (L6R1) 5


could cause a wicket. Again, the wicket rate can therefore be expressed as a function of the
parameters:

𝑑𝑊
= 𝑣(𝑊, 𝑅, 𝑂)
𝑑𝑂

Knowledge of the run rate and wicket rate at all stages in the innings should theoretically
determine the final score of a team after its full quota of overs, a par score required for the team
batting second to be on to determine the result of the match if rain stops play, and also set an
adjusted total if a section of the batting innings is lost due to rain.

Heat Equations & Fourier Transforms

Although it is useful to understand that the three parameters affect the run rate and the wicket
rate, it would be more helpful if this could be quantified. One way this could be done is by
seeing how 𝑊 and 𝑅 affect the run rate and wicket rate as 𝑂 changes, essentially turning this
into a two-variable problem. At the end of an innings, teams are more likely to try to boost their
run rates without taking 𝑊 and 𝑅 into account, as they look to maximise the potential of their
remaining resources. Because of this, the effect of 𝑊 and 𝑅 evens out as time increases.

Due to the useful property that 𝑂 is a representation of time, this can be compared to the heat
distribution of a 2-D plane. At time 𝑡 = 0, the 2-D plane has an unequal heat distribution, and
over time, the thermal energy in hotter areas of the plane will be transferred to cooler areas,
and so the heat distribution of the plane will eventually become uniform. Similarly, the
variables 𝑊 and 𝑅 could be used to identify co-ordinates on the 2-D object, and the temperature
of the object at the co-ordinates (𝑊, 𝑅) could represent the run rate at that point. The heat
distribution of the plane evening out over time is comparable to the reducing impact of 𝑊 and
𝑅 as 𝑂 increases.

This has now become a heat equation, which can be solved using a Fourier transform. This
means that the run rate at any point in a match can be predicted using the two parameters 𝑊
12 14
and 𝑅, given a function relating 13, 𝑊 and 𝑅. Similarly, a function relating 13
, 𝑊 and 𝑅 can

Sai Mehta (L6R1) 6


be formulated using the same process, again because the effects of 𝑊 and 𝑅 even out as 𝑂
increases.

Solving the Heat Equation for Run Rate

The function 𝑢(𝑊, 𝑅, 𝑂) denotes the temperature of the plane after a time 𝑂 at the co-ordinates
(𝑊, 𝑅). To perform the heat equation, conditions must be set for the plane whose temperature
is being recorded at all times. One of these is that the plane is rectangular, and all 𝑊 inputs
must fall between 0 and 10, as these are the only numbers of wickets that can be lost. All 𝑅
inputs must fall between 0 and 𝑇, where 𝑇 is the target set for the chasing team. This is because
once 𝑅 = 𝑇 the match is officially over, and no more runs can be scored. Another condition is
12
that at the boundary 𝑊 = 10, 13 must be equal to 0, and so it can be said that 𝑢(10, 𝑅, 𝑂) = 0.

This is similar to the boundary at 𝑅 = 𝑇, as no more runs can be scored and so it can be said
12
that 13
must also equal 0 here. As a result, the value for 𝑢(𝑊, 𝑇, 𝑂) = 0. Since there is a

minimum-overs threshold which must be met in the second innings, the run rate when 𝑅 = 0
is known, and so the value for 𝑢(𝑊, 0, 𝑂) is also known, which can be set to some constant
𝑘2# . The same can be done for 𝑢(0, 𝑅, 𝑂), and so this can be set to some constant 𝑘4# . This
could be similar to the effects of an insulating material being put in place on these edges of the
plane. Knowledge of the material of the plane is also of use because the conductivity of the
plane affects the temperature distribution throughout the plane.

Given a function 𝑓(𝑊, 𝑅) such that 𝑢(𝑊, 𝑅, 0) = 𝑓(𝑊, 𝑅), it is now possible to solve the heat
equation, depending on the type of function 𝑓(𝑊, 𝑅) is. The thought process behind the
solution to the heat equation, is considering the changes in temperature of each individual
particle within the plane because of the differences between its temperature and the
temperatures of the surrounding particles. If the temperature of a specific particle is less than
the average temperature of the surrounding particles, the temperature of that particle will rise
due to conduction. The increase in temperature over time can be modelled as the first derivative
of 𝑢 with respect to 𝑂 and is proportional to the difference between the mean of the
temperatures surrounding particles and the particle itself, which is essentially the second
derivative of 𝑢 with respect to the position of the particle, denoted by the co-ordinates (𝑊, 𝑅).
The boundary conditions here are important because otherwise this model would assume that

Sai Mehta (L6R1) 7


heat would be transferred out of the plane, which would result in thermal energy dissipating to
the surroundings.

In general, the partial differential equation for a problem like this is:

-
𝜕𝑢 𝜕"𝑢
= 𝛼H "
𝜕𝑡 𝜕𝑥5
56!

for an 𝑛-dimensional object, where 𝑡 is the time passed, 𝛼 is some proportionality constant, 𝑢
is the temperature and 𝑥5 denotes a particular dimension.

Figure 3 below is a graph showing the temperature of a rod of length 2𝜋 after a certain period
of time, where 𝑥 is the relative position on the rod. The rod can be modelled as a 1-D object,
and so the following formula can be used to calculate the heat distribution of the rod at a given
time.

𝜕𝑢 𝜕"𝑢
= 𝛼 ",
𝜕𝑡 𝜕𝑥

where 𝑥 signifies a particular position on the rod (Hancock, 2006).

Sai Mehta (L6R1) 8


Figure 3: Visualisation of Heat Equation (Sanderson, 2019)

As can be seen on the graph, the temperature of the rod remains unchanged at the extreme
insulated ends of the rod, and everywhere else in the rod the temperature evens out over time.
Visualisations of the 2-D model used for calculating run rate would require a 4-variable graph
including 𝑅, 𝑊, 𝑂 and 𝑢 and so this cannot be visualised in the same way, unlike the
aforementioned methods devised by Duckworth, Lewis and Stern and Jayadevan.

Since the object being observed is 2-dimensional, the following general formula to this problem
can be applied:

𝜕𝑢 𝜕"𝑢 𝜕"𝑢
= 𝛼! K + L
𝑑𝑂 𝜕𝑊 " 𝜕𝑅"

Here, 𝛼! is the diffusion constant, which shows how quickly the heat spreads within the 2-D
plane.

Sai Mehta (L6R1) 9


12
It is now possible to derive the function 𝑢(𝑊, 𝑅, 𝑂) for 13
by solving the above partial

differential equation. If 𝑢(𝑊, 𝑅, 𝑂) = 𝑋(𝑊)𝑌(𝑅)𝑍(𝑂), then from the heat equation it can be
deduced that:
𝑍 7 (𝑂) 𝑋 77 (𝑊) 𝑌 77 (𝑅)
=K + L
𝛼! 𝑍(𝑂) 𝑋(𝑊) 𝑌(𝑅)

Since the left-hand side of the equation only relies on 𝑂 and the right-hand side only relies on
8 $ (3) < $$ (4) = $$ (2) 8 $ (3) < $$ (4) = $$ (2)
𝑊 and 𝑅, let ; 8(3)
= O <(4) + =(2)
P = −𝜆" . Equating ; 8(3)
and O <(4) + =(2)
P to some
# #
%
constant −𝜆" , it can be concluded that 𝑍(𝑂) = 𝐴𝑒 %;# / 3 , 𝑋(𝑊) = 𝐵 cos(𝜆𝑊) + 𝐶 sin(𝜆𝑊)
and 𝑌(𝑅) = 𝐷 cosh[𝜆(𝑇 − 𝑅)\ + 𝐸 sinh[𝜆(𝑇 − 𝑅)\, where 𝐴, 𝐵, 𝐶, 𝐷 and 𝐸 are constants,
whose values can be determined with the aforementioned boundary conditions. We have that
𝑋(0) = 𝐵 = 𝑘4# and 𝑋(10) = 𝐵 cos(10𝜆) + 𝐶 sin(10𝜆) = 0, so 𝐶 = −𝐵 cot(10𝜆),
meaning that 𝑋(𝑊) = 𝑘4# cos(𝜆𝑊) − 𝑘4# cot(10𝜆) sin(𝜆𝑊). We also have that 𝑌(𝑇) =
𝐷 = 0 and 𝑌(0) = 𝐷 cosh(𝜆𝑇) + 𝐸 sinh(𝜆𝑇) = 𝑘2# , so 𝐸 = −𝐷 coth(𝜆𝑇), meaning that
𝑌(𝑅) = 𝑘2# cosh[𝜆(𝑇 − 𝑅)\ − 𝑘2# coth(𝜆𝑇) sinh[𝜆(𝑇 − 𝑅)\. We can conclude that:

(
#
𝑢(𝑊, 𝑅, 𝑂) = ) 𝐴! 𝑒 "#!$"% ,𝑘&! cos(𝜆! 𝑊) − 𝑘&! cot(10𝜆! ) sin(𝜆! 𝑊)8,𝑘'! cosh,𝜆! (𝑇 − 𝑅)8 − 𝑘'! coth(𝜆! 𝑇) sinh,𝜆! (𝑇 − 𝑅)88
!)*

where 𝐴- and 𝜆- are constants that vary for different values of 𝑛.

Solving the Heat Equation for Wicket Rate

A similar calculation can be made for both the run rate and wicket rate, as both variables rely
on the same assumption that the effects of 𝑅 and 𝑊 even out as 𝑂 increases. The function
14
𝑣(𝑊, 𝑅, 𝑂) can be used to denote 13
after time 𝑂, using the temperature distribution of another
12 14
2-D plane to calculate this. Just as 13 must be 0 after a team has been bowled out, 13
= 0 when
14
𝑊 = 10 and so 𝑣(10, 𝑅, 𝑂) = 0. We can also assume that 13
is equal to 0 when the innings

has finished, and so 𝑣(𝑊, 𝑇, 𝑂) = 0 as no further wickets can be lost after the chasing side has
reached the target score. Then, the values of 𝑣(𝑊, 0, 𝑂) and 𝑣(0, 𝑅, 𝑂) can be 𝑘2% and 𝑘4%

Sai Mehta (L6R1) 10


12
respectively. Since similar values have been chosen for the same inputs as the function for 13,

a very similar formula for 𝑣(𝑊, 𝑅, 𝑂) can be produced as 𝑢(𝑊, 𝑅, 𝑂), that is:

(
#
𝑣(𝑊, 𝑅, 𝑂) = ) 𝐵! 𝑒 "##+"% ,𝑘&# cos(𝜇! 𝑊) − 𝑘&# cot(10𝜆! ) sin(𝜆! 𝑊)8,𝑘'# cosh,𝜆! (𝑇 − 𝑅)8 − 𝑘'# coth(𝜆! 𝑇) sinh,𝜆! (𝑇 − 𝑅)88
!)*

where 𝛼" is the diffusion constant and 𝐵- and 𝜇- are constants that vary for different values of
𝑛.

Calculating the Initial Conditions

The formulae to determine both the run rate and wicket rate require constant values. It could
12 14
be that the values of the diffusion gradients, which show how much 𝑅 and 𝑊 affect 13 and 13
,

are given for each ground or weather conditions. For example, in lower-scoring grounds such
as the BRSABV Ekana Stadium, Lucknow, batsmen will be less incentivised to take risks and
will take longer to acclimatise to conditions, as the pitch is generally slow, so the diffusion
gradient for run rate will be relatively large – losing a wicket should have a large impact on the
rate at which runs are scored.

However, there would be a small sample size due to the large number of grounds used
internationally. One example of this is that 53 stadiums have been used internationally in India.
Since the majority of international matches in India are bilateral series between India and
another touring country, some grounds are used very rarely. While Eden Gardens, Kolkata, has
hosted 31 ODIs and 11 T20Is, the Shaheed Veer Narayan Singh International Cricket Stadium,
Raipur, has only hosted 1 ODI, which was played on 21st January 2023, and no T20Is as of
12 14
June 2023. Therefore, calculating the constants required for the 13 and 13
equations using this

method would be very unreliable.

As an alternative, neural networks could be used to calculate the constants. By training a


computer on over 4500 ODI and 2000 T20I matches where constants can be determined in
hindsight, it will detect patterns within the data, so that if a match is interrupted with rain, the
computer can then use data from the match to calculate the constants. This means that matches
on grounds with less data can have more reliable values for the constants.

Sai Mehta (L6R1) 11


Using Run Rate & Wicket Rate to Predict Final Score

12 14
One drawback of the method proposed above is that the functions for 13 and 13
rely on 𝑅 and

𝑊 themselves. This means that finding 𝑅 and 𝑊when 𝑂 is equal to the number of allowed
overs in an innings cannot be calculated through using simple integration techniques.

One way of getting around this is the use of a Runge-Kutta method, which can be used to
numerically solve the presented differential equation. The most common Runge-Kutta method
is the RK4 method, which is a fourth order Runge-Kutta equation.

1>
If 1?
= 𝑓(𝑥, 𝑦) and 𝑦(𝑥# ) = 𝑦# where the function 𝑓 is known, and the initial values of 𝑥 and

𝑦, 𝑥# and 𝑦# respectively, with some step ℎ, then:

𝑥-.! = 𝑥- + ℎ

𝑦-.! = 𝑦- + (𝑘! + 2𝑘" + 2𝑘0 + 𝑘@ ),
6

(Süli & Mayers, 2003)


for 𝑛 ∈ ℤ. , where:

𝑘! = 𝑓(𝑥- , 𝑦- )
ℎ 𝑘!
𝑘" = 𝑓 d𝑥- + , 𝑦- + ℎ f
2 2
ℎ 𝑘"
𝑘0 = 𝑓 d𝑥- + , 𝑦- + ℎ f
2 2
𝑘@ = 𝑓(𝑥- + ℎ, 𝑦- + ℎ𝑘0 )

The RK4 method approximates the value of 𝑦-.! = 𝑦(𝑥-.! ) using the previous value of 𝑦-
and the weighted average of the four increments 𝑘! , 𝑘" , 𝑘0 , 𝑘@ , which are values of the function
taken at intervals of different multiples of ℎ. This is the most commonly used Runge-Kutta
method because it provides a reasonably high accuracy and although using higher order
methods may reduce error, they require more calculations, and so a fourth order equation is
preferred.

Sai Mehta (L6R1) 12


In the case of DLS, the Runge-Kutta method can be used to calculate 𝑅 and 𝑊 using their
derivatives with respect to 𝑂. Since the functions 𝑢 and 𝑣 have been calculated by solving the
heat equation, the RK4 method can be used to find 𝑅 and 𝑊 for each value of 𝑂 by using an
!
increment ℎ. This increment could be ℎ = A to represent one ball or ℎ = 1 to represent an over.

This choice would depend on the time taken for the calculations to be made, as so may differ
across different formats of the game. For example, if only 20 overs have been bowled in a One
Day International, the minimum for the result of a match to be decided by DLS, then 30 overs
!
still remain and using ℎ = A may mean the process will take too long. On the other hand, this

may not be as much of a problem for a shorter format match such as T10.

The advantage of using a method like this is that it can calculate the par score a chasing side
should be on when chasing a total. This means that if a match is to be abandoned due to rain,
the result of the match can be determined and quantified by comparing the par score and the
current score the batting side is on. If rain delays play at any stage in the match, such that play
may resume at a later stage in the match, this method can be used to calculate how many runs
the batting team would have scored, so that an adjusted target can be set for the team batting
second to chase down.

Limitations of the Proposed Model

One problem with the proposed model is the existence of varying field restrictions within
limited overs formats of cricket. For example, in most T20 matches, only two fielders may be
allowed outside the 30-yard circle for the first 6 overs of the match, but after this, bowling sides
are allowed to place five fielders outside the 30-yard circle. This can be seen from Figure 2,
showing that batting teams tend to bat slower during the middle phases of the game, after the
field restrictions are relaxed. While the underlying assumption of the model, that the effects of
𝑊 and 𝑅 even out over time, still holds, we see how external factors can also have a large
12 14
impact on 13
and 13
. However, this problem can be mitigated by using different diffusion

constants, 𝛼, for the three phases of the game - the powerplay (overs 1 to 6), the middle overs
(overs 7 to 15) and the death overs (overs 16 to 20) in a T20. Teams are usually very willing to
12 14
play aggressively in the powerplay and so the effects of 𝑊 and 𝑅 on 13 and 13
are much less,

Sai Mehta (L6R1) 13


meaning that the diffusion constant in this scenario should be less than for the middle overs.
The values of 𝛼 can be calculated for each phase, rather than for the whole game, which not
only mitigates the problem of fielding restrictions, but also accounts for changing playing
conditions which could change, especially given that this method is used for rain-affected
matches.

Another potential limitation of the proposed model is the evolution of cricket. One problem for
any model for rain-affected cricket matches is that general tactics tend to change over time.
One example of this is the addition of the Impact Player rule in the TATA Indian Premier
League 2023, which meant teams could make a permanent substitution at any stage in a match,
effectively playing 12 players rather than 11. Figure 4 shows the effect of this on scoring rates.

Figure 4: Scoring patterns in first innings totals in the IPL (Brar, 2023)

From the above chart, it can be seen that 54% of first innings totals were 180 or more in 2023,
but the next highest proportion of scores of 180 or more in any previous season was 37% in
2015. Also, teams ‘scored 200 or more 37 times in 2023 – more than double the previous record
of 18’, set in 2022 (Brar, 2023). The Impact Player rule meant that teams could be more
aggressive, knowing that they could utilise the one extra batsman, and so the effect of losing a
wicket was much less. With cricket tournaments introducing their own regulations, another
being the Super Surge in the Big Bash League, and teams using different tactics to adapt to
them, changes must be made to the initial conditions of the model. There could be different

Sai Mehta (L6R1) 14


initial conditions used for different tournaments such as the IPL and BBL, or even new
conditions produced each year to combat the evolution of tactics and mindset across all
tournaments – for example, it is becoming a general trend globally that spinners are bowling
much more overs outside of the middle overs.

Conclusion

In this project, three methods have been viewed to tackle the problem of cricket matches
interrupted by rain – the Duckworth Lewis Stern method based on the batting team’s remaining
resources, V. Jayadevan’s method based on regression to derive an equation for percentage of
runs scored, and the proposed model in the project that involves solving a 2-D heat equation to
calculate the derivatives of runs and wickets with respect to overs bowled.

The aim of this project was to evaluate the DLS method and to seek an alternative solution to
the problem.

The primary differences between the above three approaches are the underlying assumptions
that are used to help reach an answer – one assumption being that runs follow an exponential
model, another a cubic model determined by accelerations in the innings, and the third, that
runs and wickets have a declining effect on the run rate and wicket rate as the innings
progresses.

Although a potential solution has been proposed in this project, applying a heat equation, as a
clear limitation, from a practical perspective it remains to be tested. Subsequent to the
application of the proposed to determine its suitability, it may require further adjustment to
enable the attainment of a fair and accurate result for rain-affected cricket matches.

Overall, the main problem with all three approaches is that the assumptions behind them are
not completely accurate, culminating in slight inaccuracies which in turn have ultimately
resulted in numerous controversies. It is therefore unlikely that it will be possible to create a
model that can provide a consistent and perfect outcome.

Sai Mehta (L6R1) 15


Bibliography

Brar, H., 2023. IPL 2023 takeaways: Runs get quicker and bigger as Impact Players have their
say. ESPN cricinfo, 31 May.
Duckworth, F. C. & Lewis, A. J., 2004. A Successful Operational Research Intervention in
One-Day Cricket. The Journal of Operational Research Society, July, 55(7), pp. 749-759.
Hancock, M. J., 2006. The 1-D Heat Equation, s.l.: MIT.
Jayadevan, V., 2002. A new method for the computation of target scores in interrupted, limited-
over cricket matches. Current Science, 83(5), pp. 577-586.
Süli, E. & Mayers, D., 2003. An Introduction to Numerical Analysis. IBSN 0-521-00794-1 ed.
s.l.:Cambridge University Press.
Sanderson, G., 2019. Solving the heat equation | DE3. [Online]
Available at: https://www.youtube.com/watch?v=ToIXSwZ1pJU
Stern, S. E., 2016. The Duckworth-Lewis-Stern method: extending the Duckworth-Lewis
methodology to deal with modern scoring rates. Journal of the Operational Research Society,
December, 67(12), pp. 1469-1480.

Sai Mehta (L6R1) 16

You might also like