Authors

  • Jonqobilov Jahongir Tirkashevich

DOI:

https://doi.org/10.71337/inlibrary.uz.wsrj.100613

Keywords:

Keywords: differential equations first-order numerical methods Runge-Kutta machine learning MATLAB Python.

Abstract

Abstract: First-order differential equations play a crucial role in mathematics, physics, engineering, and economics. This article examines modern methods for solving first-order differential equations, focusing on numerical techniques (e.g., Euler, Runge-Kutta), computational tools (e.g., MATLAB, Python), and machine learning-based approaches. The advantages, limitations, and practical applications of these methods are discussed. The aim of this study is to compare these methods, evaluate their effectiveness, and provide guidance for researchers and practitioners.


background image

World scientific research journal

https://scientific-jl.com/wsrj

Volume-40_Issue-1_June-2025

100

MODERN METHODS FOR SOLVING FIRST-ORDER

DIFFERENTIAL EQUATIONS

Jonqobilov Jahongir Tirkashevich

Tashkent State Technical University,

Almalyk Branch, Assistant

Abstract:

First-order differential equations play a crucial role in mathematics,

physics, engineering, and economics. This article examines modern methods for
solving first-order differential equations, focusing on numerical techniques (e.g.,
Euler, Runge-Kutta), computational tools (e.g., MATLAB, Python), and machine
learning-based approaches. The advantages, limitations, and practical applications of
these methods are discussed. The aim of this study is to compare these methods,
evaluate their effectiveness, and provide guidance for researchers and practitioners.

Keywords:

differential equations, first-order, numerical methods, Runge-Kutta,

machine learning, MATLAB, Python.

Introduction

Differential equations serve as a cornerstone for modeling dynamic processes

in natural and social sciences. First-order differential equations, expressed in the
general form dy/dx=f(x,y) describe the rate of change of a function y(x) y(x) y(x)
with respect to an independent variable x x x. These equations are ubiquitous in
applications such as physics (e.g., Newton’s laws of motion), biology (e.g., population
dynamics), engineering (e.g., control systems), and economics (e.g., financial
forecasting). Classical analytical methods, such as separation of variables, integrating
factor techniques, and exact equations, are effective for solving simple linear or
separable first-order differential equations. However, many real-world problems
involve nonlinear, stiff, or complex equations that are intractable analytically. The
advent of modern computational techniques has transformed the field, enabling
accurate and efficient solutions through numerical methods, computational software,
and innovative machine learning approaches. This article provides an in-depth
exploration of modern methods for solving first-order differential equations,
including numerical algorithms (Euler, Improved Euler, and Runge-Kutta),
computational tools (MATLAB, Python, and Julia), and machine learning techniques
(physics-informed neural networks). The study aims to compare these methods in
terms of accuracy, computational cost, and applicability, offering a comprehensive
resource for researchers, engineers, and scientists. Additionally, it discusses
challenges, limitations, and future directions for solving first-order differential
equations in increasingly complex systems.


background image

World scientific research journal

https://scientific-jl.com/wsrj

Volume-40_Issue-1_June-2025

101

Literature Review

The development of methods for solving first-order differential equations has

evolved significantly over centuries. Classical analytical techniques, pioneered by
mathematicians like Euler, Bernoulli, and Lagrange in the 18th and 19th centuries,
laid the foundation for solving linear and separable equations (Boyce & DiPrima,
2012). Methods such as separation of variables and the integrating factor technique
are effective for equations with closed-form solutions but falter when applied to
nonlinear or complex systems.

In the 20th century, numerical methods emerged to address these limitations.

The Euler method, one of the earliest numerical techniques, approximates solutions
by discretizing the differential equation (Butcher, 2008). While simple, its low
accuracy spurred the development of more sophisticated methods, such as the Runge-
Kutta family, which offers higher-order accuracy through iterative approximations
(Runge & Kutta, 1895). The fourth-order Runge-Kutta (RK4) method, in particular,
became a standard due to its balance of accuracy and computational efficiency.

The rise of computational tools in the late 20th and early 21st centuries

revolutionized numerical solutions. Software like MATLAB, Python (via libraries
such as scipy.integrate), and Julia (via DifferentialEquations.jl) has made numerical
methods accessible and efficient for large-scale problems (Chapra & Canale, 2015).
These tools automate complex calculations and provide built-in solvers optimized for
various equation types, including stiff systems.

More recently, machine learning has introduced novel approaches to solving

differential equations. Physics-informed neural networks (PINNs), proposed by
Raissi et al. (2019), leverage deep learning to approximate solutions by embedding
the governing equations into the neural network’s loss function. This approach is
particularly promising for high-dimensional, nonlinear, or stiff systems where
traditional methods may struggle. Other machine learning techniques, such as
Gaussian processes and reinforcement learning, are also being explored for specific
applications (Lagaris et al., 1998). This literature review synthesizes classical,
numerical, computational, and machine learning-based approaches, providing a
foundation for analyzing modern methods and their practical implications.

Literature Review

The development of methods for solving first-order differential equations has

evolved significantly over centuries. Classical analytical techniques, pioneered by
mathematicians like Euler, Bernoulli, and Lagrange in the 18th and 19th centuries,
laid the foundation for solving linear and separable equations (Boyce & DiPrima,
2012). Methods such as separation of variables and the integrating factor technique
are effective for equations with closed-form solutions but falter when applied to
nonlinear or complex systems. In the 20th century, numerical methods emerged to
address these limitations. The Euler method, one of the earliest numerical techniques,


background image

World scientific research journal

https://scientific-jl.com/wsrj

Volume-40_Issue-1_June-2025

102

approximates solutions by discretizing the differential equation (Butcher, 2008).
While simple, its low accuracy spurred the development of more sophisticated
methods, such as the Runge-Kutta family, which offers higher-order accuracy through
iterative approximations (Runge & Kutta, 1895). The fourth-order Runge-Kutta
(RK4) method, in particular, became a standard due to its balance of accuracy and
computational efficiency. The rise of computational tools in the late 20th and early
21st centuries revolutionized numerical solutions. Software like MATLAB, Python
(via libraries such as scipy.integrate), and Julia (via DifferentialEquations.jl) has
made numerical methods accessible and efficient for large-scale problems (Chapra &
Canale, 2015). These tools automate complex calculations and provide built-in
solvers optimized for various equation types, including stiff systems.More recently,
machine learning has introduced novel approaches to solving differential equations.
Physics-informed neural networks (PINNs), proposed by Raissi et al. (2019), leverage
deep learning to approximate solutions by embedding the governing equations into
the neural network’s loss function. This approach is particularly promising for high-
dimensional, nonlinear, or stiff systems where traditional methods may struggle.
Other machine learning techniques, such as Gaussian processes and reinforcement
learning, are also being explored for specific applications (Lagaris et al., 1998). This
literature review synthesizes classical, numerical, computational, and machine
learning-based approaches, providing a foundation for analyzing modern methods
and their practical implications.

Modern Methods

1. Numerical Methods
Numerical methods approximate solutions to first-order differential equations by

discretizing the continuous problem into a series of iterative steps. For an equation of
the form dy/dx=f(x,y) with initial condition y(x

0

)=y

0

, numerical methods compute

approximate values y

n

≈y(x

n

) at discrete points x

n

=x

0

+nh, where h h h is the step size.

Below, we discuss key numerical methods in detail.

1.1 Euler Method

The Euler method is the simplest numerical approach, approximating the

solution using the formula:

𝑦

𝑛+1

= 𝑦

𝑛

+ ℎ𝑓(𝑥

𝑛

, 𝑦

𝑛

)

1.


This method uses the tangent line at each point to estimate the next value. While

computationally inexpensive, its local truncation error is O(h

2

), leading to significant

inaccuracies for small step sizes or stiff equations. The Euler method is best suited for
simple problems or educational purposes but is rarely used in practice for complex
systems due to its low accuracy.


background image

World scientific research journal

https://scientific-jl.com/wsrj

Volume-40_Issue-1_June-2025

103

1.2 Improved Euler Method (Heun’s Method)

The Improved Euler method, also known as Heun’s method, enhances the basic

Euler approach by incorporating a predictor-corrector strategy. It computes an initial
estimate (predictor) using the Euler method and then refines it (corrector) by
averaging

the

slopes

at

the

current

and

predicted

points:

𝑦

𝑛+1

𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑜𝑟

= 𝑦

𝑛

+ ℎ𝑓(𝑥

𝑛

, 𝑦

𝑛

)

2.

This method reduces the local truncation error to O(h

3

), offering improved accuracy

over the Euler method with only a modest increase in computational cost.

1.3 Runge-Kutta Methods

The Runge-Kutta (RK) family of methods provides higher-order

approximations by evaluating the function f(x,y) f(x, y) f(x,y) at multiple intermediate
points within each step. The fourth-order Runge-Kutta (RK4) method is particularly
popular due to its accuracy and efficiency. The RK4 algorithm is given by:

𝑘

1

= 𝑓(𝑥

𝑛

, 𝑦

𝑛

)

𝑘

2

= 𝑓(𝑥

𝑛

+


2

, 𝑦

𝑛+


2

𝑘

1

)

𝑘

3

= 𝑓(𝑥

𝑛

+

2

, 𝑦

𝑛+

2

𝑘

2

)

3.

𝑘

4

= 𝑓(𝑥

𝑛+1

, 𝑦

𝑛

+ ℎ𝑘

3

)

𝑦

𝑛+1

= 𝑦

𝑛

+


6

(𝑘

1

+ 2𝑘

2

+ 3𝑘

3

+ 𝑘

4

)

With a local truncation error of O(h

5

), RK4 is highly accurate and widely used

in applications requiring precise solutions, such as orbital mechanics and fluid
dynamics.

1.4 Methods for Stiff Equations

Stiff differential equations, characterized by rapidly changing solutions or

widely varying time scales, pose challenges for explicit methods like Euler or RK4.
Implicit methods, such as the Backward Euler method or the Trapezoidal Rule, are
often employed for stiff systems. For example, the Backward Euler method solves:

𝑦

𝑛+1

= 𝑦

𝑛

+ ℎ𝑓(𝑥

𝑛

, 𝑦

𝑛

)

4.

This requires solving a nonlinear equation at each step, typically using iterative
techniques like Newton’s method. Implicit methods are computationally intensive but
stable for stiff problems, making them suitable for applications like chemical kinetics
or electrical circuits.

2. Computational Tools: Modern computational tools have transformed the

solution of differential equations by automating numerical methods and enabling
large-scale simulations. Below, we discuss three prominent tools: MATLAB, Python,
and Julia.


background image

World scientific research journal

https://scientific-jl.com/wsrj

Volume-40_Issue-1_June-2025

104

2.1 MATLAB

MATLAB is a widely used platform for numerical computations, offering

robust solvers for differential equations. The ode45 function, based on an adaptive
Runge-Kutta method (Dormand-Prince pair), automatically adjusts the step size to
balance accuracy and efficiency. For stiff equations, MATLAB provides solvers like
ode15s, which uses implicit methods. MATLAB’s user-friendly interface and
visualization capabilities make it ideal for engineering and scientific applications.

2.2 Python

Python’s scipy.integrate module provides powerful tools for solving

differential equations. The odeint function, based on the LSODA algorithm,
automatically switches between explicit and implicit methods depending on the
equation’s stiffness. The following example demonstrates solving the equation
dy/dx=−2xy with initial condition y(0)=1:

from

scipy.integrate

import

odeint

import

numpy

as

np

import

matplotlib.pyplot

as

plt

def

model

(y, x):

return

-

2

* x * y

# dy/dx = -2xy

x = np.linspace(

0

,

5

,

100

)

y0 =

1

y = odeint(model, y0, x)
plt.plot(x, y, label=

"Solution"

)

plt.xlabel(

"x"

)

plt.ylabel(

"y"

)

plt.title(

"Solution of dy/dx = -2xy"

)

plt.legend()
plt.grid(

True

)

plt.show()
This code produces a numerical solution and visualizes it, demonstrating

Python’s versatility for both computation and visualization.

2.3 Julia

Julia, a newer programming language, is gaining popularity for scientific

computing due to its high performance. The DifferentialEquations.jl package offers a
comprehensive suite of solvers for differential equations, including adaptive Runge-
Kutta methods, implicit methods for stiff systems, and specialized algorithms for
stochastic or delay differential equations. Julia’s speed and flexibility make it a
compelling choice for large-scale simulations.


background image

World scientific research journal

https://scientific-jl.com/wsrj

Volume-40_Issue-1_June-2025

105

3. Machine Learning Approaches
Machine learning has introduced innovative methods for solving differential

equations, particularly for complex or high-dimensional systems. Physics-informed
neural networks (PINNs) are a prominent example, combining deep learning with
physical constraints to approximate solutions.

3.1 Physics-Informed Neural Networks (PINNs)

PINNs embed the differential equation and its boundary or initial conditions

into the loss function of a neural network. For a first-order differential equation
dy/dx=f(x,y), the neural network approximates y(x) as a function parameterized by
weights and biases. The loss function includes terms for the differential equation
residual, initial conditions, and boundary conditions (if applicable). For example, the
loss

function

for

the

equation

dy/dx=−2xy

might

be:

Loss=∑(dy/dx+2xy)

2

+(y(0)−1)

2

PINNs are particularly effective for nonlinear equations, high-dimensional systems,
or problems with irregular domains, where traditional methods may struggle.

3.2 Other Machine Learning Approaches

Beyond PINNs, other machine learning techniques are being explored.

Gaussian processes can model uncertainty in solutions, while reinforcement learning
has been applied to optimize numerical solvers. These methods are still in early stages
but show promise for specialized applications, such as inverse problems or parameter
estimation.

Applications

Modern methods for solving first-order differential equations have broad

applications across disciplines:

Physics

: Modeling motion, heat transfer, and electromagnetic systems (e.g.,

solving dv/dt=−g−(k/m)v for a falling object with air resistance).

Economics

: Forecasting financial markets or modeling economic growth (e.g.,

Solow-Swan growth models).

Environmental Science

: Modeling climate systems or pollutant dispersion.

For example, the Runge-Kutta method is used in orbital mechanics to predict

satellite trajectories, while PINNs have been applied to fluid dynamics problems with
complex geometries. Computational tools like MATLAB and Python enable rapid
prototyping and visualization, making them indispensable in research and industry.

Discussion

The modern methods discussed offer distinct advantages and limitations:

Numerical Methods

: The Euler method is computationally simple but lacks

accuracy, making it suitable only for basic problems. The Improved Euler and Runge-
Kutta methods offer higher accuracy but require more computational resources.
Implicit methods excel for stiff equations but involve complex iterative solutions.


background image

World scientific research journal

https://scientific-jl.com/wsrj

Volume-40_Issue-1_June-2025

106

Computational Tools

: MATLAB, Python, and Julia streamline numerical

computations and provide adaptive solvers for various equation types. MATLAB is
user-friendly but proprietary, while Python and Julia are open-source and highly
customizable. Julia’s performance advantages make it ideal for large-scale problems.

Machine Learning

: PINNs provide flexibility for nonlinear and high-

dimensional systems but require significant computational power and expertise in
deep learning. They are less mature than numerical methods but show promise for
future applications.

Challenges include balancing accuracy and computational cost, handling stiff

equations, and scaling methods to high-dimensional systems. For instance, while RK4
is accurate for smooth solutions, it may fail for stiff problems, necessitating implicit
methods. Similarly, PINNs require careful tuning of neural network architectures and
large datasets for training. Future directions include hybrid approaches that combine
numerical methods with machine learning, such as using neural networks to optimize
step sizes in Runge-Kutta methods. Advances in quantum computing may also enable
faster solutions for large-scale differential equations.

Conclusion

Modern methods for solving first-order differential equations—numerical

techniques, computational tools, and machine learning approaches—have
significantly advanced the ability to model complex systems. Numerical methods like
Runge-Kutta provide high accuracy for smooth problems, while implicit methods
address stiff systems. Computational tools like MATLAB, Python, and Julia automate
and optimize solutions, making them accessible to a wide audience. Machine learning,
particularly PINNs, offers innovative solutions for nonlinear and high-dimensional
problems. By understanding the strengths and limitations of these methods,
researchers and practitioners can select the most appropriate approach for their needs.
Future advancements in hybrid methods, artificial intelligence, and computational
hardware promise to further enhance the field, opening new possibilities for solving
complex differential equations.

References:

1.

Reddy, J. N.

- "An Introduction to the Finite Element Method," McGraw-Hill,

2006. Finite element usuli asoslari va ularning muhandislikda qo‘llanilishi.

2.

Kreyszig, E.

- "Advanced Engineering Mathematics," Wiley, 2011. Zamonaviy

matematik usullar va ularning muhandislik sohalaridagi qo‘llanilishi.

3.

Boyce, W. E., DiPrima, R. C.

- "Elementary Differential Equations and

Boundary Value Problems," John Wiley & Sons, 2017. Differensial tenglamalar
va chegaraviy qiymat masalalarini yechishning asosiy usullari haqida.

4.

Jonqobilov, J.T.. Texnologik Jarayonlarni Monitoring Qilish Va Vizualizatsiya
Usullari. 192-201


background image

World scientific research journal

https://scientific-jl.com/wsrj

Volume-40_Issue-1_June-2025

107

5.

Manshurov, Sh.T.; Jonqobilov, J.T.. C++ Dasturlarlash Tilida n-Xonali
Palindromik Sonlarni Topish. 173-178

6.

Djabbarov,

Odil

Djurayevich;

Jonqobilov,

Jahongir

Tirkashevich.

TRIGONOMETRIK FUNKSIYALARNI EKVIVALENT TA’RIFI HAQIDA.
581-585

7.

Муминов, Ф. М., Душатов, Н. Т., Миратоев, З. М., & Ибодуллаева, М. Ш.
ОБ ОДНОЙ КРАЕВОЙ ЗАДАЧЕ ДЛЯ УРАВНЕНИЯ ТРЕТЬЕГО
ПОРЯДКА СМЕШАННО-СОСТАВНОГО ТИПА. Innovative, educational,
natural and social sciences, 2(6), 606-612.



References

Reddy, J. N. - "An Introduction to the Finite Element Method," McGraw-Hill, 2006. Finite element usuli asoslari va ularning muhandislikda qo‘llanilishi.

Kreyszig, E. - "Advanced Engineering Mathematics," Wiley, 2011. Zamonaviy matematik usullar va ularning muhandislik sohalaridagi qo‘llanilishi.

Boyce, W. E., DiPrima, R. C. - "Elementary Differential Equations and Boundary Value Problems," John Wiley & Sons, 2017. Differensial tenglamalar va chegaraviy qiymat masalalarini yechishning asosiy usullari haqida.

Jonqobilov, J.T.. Texnologik Jarayonlarni Monitoring Qilish Va Vizualizatsiya Usullari. 192-201

Manshurov, Sh.T.; Jonqobilov, J.T.. C++ Dasturlarlash Tilida n-Xonali Palindromik Sonlarni Topish. 173-178

Djabbarov, Odil Djurayevich; Jonqobilov, Jahongir Tirkashevich. TRIGONOMETRIK FUNKSIYALARNI EKVIVALENT TA’RIFI HAQIDA. 581-585

Муминов, Ф. М., Душатов, Н. Т., Миратоев, З. М., & Ибодуллаева, М. Ш. ОБ ОДНОЙ КРАЕВОЙ ЗАДАЧЕ ДЛЯ УРАВНЕНИЯ ТРЕТЬЕГО ПОРЯДКА СМЕШАННО-СОСТАВНОГО ТИПА. Innovative, educational, natural and social sciences, 2(6), 606-612.