mirror of
https://github.com/TheAlgorithms/Python.git
synced 2024-11-24 21:41:08 +00:00
bc8df6de31
* [pre-commit.ci] pre-commit autoupdate updates: - [github.com/astral-sh/ruff-pre-commit: v0.2.2 → v0.3.2](https://github.com/astral-sh/ruff-pre-commit/compare/v0.2.2...v0.3.2) - [github.com/pre-commit/mirrors-mypy: v1.8.0 → v1.9.0](https://github.com/pre-commit/mirrors-mypy/compare/v1.8.0...v1.9.0) * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci --------- Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
232 lines
6.9 KiB
Python
232 lines
6.9 KiB
Python
"""
|
|
Use the Adams-Bashforth methods to solve Ordinary Differential Equations.
|
|
|
|
https://en.wikipedia.org/wiki/Linear_multistep_method
|
|
Author : Ravi Kumar
|
|
"""
|
|
|
|
from collections.abc import Callable
|
|
from dataclasses import dataclass
|
|
|
|
import numpy as np
|
|
|
|
|
|
@dataclass
|
|
class AdamsBashforth:
|
|
"""
|
|
args:
|
|
func: An ordinary differential equation (ODE) as function of x and y.
|
|
x_initials: List containing initial required values of x.
|
|
y_initials: List containing initial required values of y.
|
|
step_size: The increment value of x.
|
|
x_final: The final value of x.
|
|
|
|
Returns: Solution of y at each nodal point
|
|
|
|
>>> def f(x, y):
|
|
... return x + y
|
|
>>> AdamsBashforth(f, [0, 0.2, 0.4], [0, 0.2, 1], 0.2, 1) # doctest: +ELLIPSIS
|
|
AdamsBashforth(func=..., x_initials=[0, 0.2, 0.4], y_initials=[0, 0.2, 1], step...)
|
|
>>> AdamsBashforth(f, [0, 0.2, 1], [0, 0, 0.04], 0.2, 1).step_2()
|
|
Traceback (most recent call last):
|
|
...
|
|
ValueError: The final value of x must be greater than the initial values of x.
|
|
|
|
>>> AdamsBashforth(f, [0, 0.2, 0.3], [0, 0, 0.04], 0.2, 1).step_3()
|
|
Traceback (most recent call last):
|
|
...
|
|
ValueError: x-values must be equally spaced according to step size.
|
|
|
|
>>> AdamsBashforth(f,[0,0.2,0.4,0.6,0.8],[0,0,0.04,0.128,0.307],-0.2,1).step_5()
|
|
Traceback (most recent call last):
|
|
...
|
|
ValueError: Step size must be positive.
|
|
"""
|
|
|
|
func: Callable[[float, float], float]
|
|
x_initials: list[float]
|
|
y_initials: list[float]
|
|
step_size: float
|
|
x_final: float
|
|
|
|
def __post_init__(self) -> None:
|
|
if self.x_initials[-1] >= self.x_final:
|
|
raise ValueError(
|
|
"The final value of x must be greater than the initial values of x."
|
|
)
|
|
|
|
if self.step_size <= 0:
|
|
raise ValueError("Step size must be positive.")
|
|
|
|
if not all(
|
|
round(x1 - x0, 10) == self.step_size
|
|
for x0, x1 in zip(self.x_initials, self.x_initials[1:])
|
|
):
|
|
raise ValueError("x-values must be equally spaced according to step size.")
|
|
|
|
def step_2(self) -> np.ndarray:
|
|
"""
|
|
>>> def f(x, y):
|
|
... return x
|
|
>>> AdamsBashforth(f, [0, 0.2], [0, 0], 0.2, 1).step_2()
|
|
array([0. , 0. , 0.06, 0.16, 0.3 , 0.48])
|
|
|
|
>>> AdamsBashforth(f, [0, 0.2, 0.4], [0, 0, 0.04], 0.2, 1).step_2()
|
|
Traceback (most recent call last):
|
|
...
|
|
ValueError: Insufficient initial points information.
|
|
"""
|
|
|
|
if len(self.x_initials) != 2 or len(self.y_initials) != 2:
|
|
raise ValueError("Insufficient initial points information.")
|
|
|
|
x_0, x_1 = self.x_initials[:2]
|
|
y_0, y_1 = self.y_initials[:2]
|
|
|
|
n = int((self.x_final - x_1) / self.step_size)
|
|
y = np.zeros(n + 2)
|
|
y[0] = y_0
|
|
y[1] = y_1
|
|
|
|
for i in range(n):
|
|
y[i + 2] = y[i + 1] + (self.step_size / 2) * (
|
|
3 * self.func(x_1, y[i + 1]) - self.func(x_0, y[i])
|
|
)
|
|
x_0 = x_1
|
|
x_1 += self.step_size
|
|
|
|
return y
|
|
|
|
def step_3(self) -> np.ndarray:
|
|
"""
|
|
>>> def f(x, y):
|
|
... return x + y
|
|
>>> y = AdamsBashforth(f, [0, 0.2, 0.4], [0, 0, 0.04], 0.2, 1).step_3()
|
|
>>> y[3]
|
|
0.15533333333333332
|
|
|
|
>>> AdamsBashforth(f, [0, 0.2], [0, 0], 0.2, 1).step_3()
|
|
Traceback (most recent call last):
|
|
...
|
|
ValueError: Insufficient initial points information.
|
|
"""
|
|
if len(self.x_initials) != 3 or len(self.y_initials) != 3:
|
|
raise ValueError("Insufficient initial points information.")
|
|
|
|
x_0, x_1, x_2 = self.x_initials[:3]
|
|
y_0, y_1, y_2 = self.y_initials[:3]
|
|
|
|
n = int((self.x_final - x_2) / self.step_size)
|
|
y = np.zeros(n + 4)
|
|
y[0] = y_0
|
|
y[1] = y_1
|
|
y[2] = y_2
|
|
|
|
for i in range(n + 1):
|
|
y[i + 3] = y[i + 2] + (self.step_size / 12) * (
|
|
23 * self.func(x_2, y[i + 2])
|
|
- 16 * self.func(x_1, y[i + 1])
|
|
+ 5 * self.func(x_0, y[i])
|
|
)
|
|
x_0 = x_1
|
|
x_1 = x_2
|
|
x_2 += self.step_size
|
|
|
|
return y
|
|
|
|
def step_4(self) -> np.ndarray:
|
|
"""
|
|
>>> def f(x,y):
|
|
... return x + y
|
|
>>> y = AdamsBashforth(
|
|
... f, [0, 0.2, 0.4, 0.6], [0, 0, 0.04, 0.128], 0.2, 1).step_4()
|
|
>>> y[4]
|
|
0.30699999999999994
|
|
>>> y[5]
|
|
0.5771083333333333
|
|
|
|
>>> AdamsBashforth(f, [0, 0.2, 0.4], [0, 0, 0.04], 0.2, 1).step_4()
|
|
Traceback (most recent call last):
|
|
...
|
|
ValueError: Insufficient initial points information.
|
|
"""
|
|
|
|
if len(self.x_initials) != 4 or len(self.y_initials) != 4:
|
|
raise ValueError("Insufficient initial points information.")
|
|
|
|
x_0, x_1, x_2, x_3 = self.x_initials[:4]
|
|
y_0, y_1, y_2, y_3 = self.y_initials[:4]
|
|
|
|
n = int((self.x_final - x_3) / self.step_size)
|
|
y = np.zeros(n + 4)
|
|
y[0] = y_0
|
|
y[1] = y_1
|
|
y[2] = y_2
|
|
y[3] = y_3
|
|
|
|
for i in range(n):
|
|
y[i + 4] = y[i + 3] + (self.step_size / 24) * (
|
|
55 * self.func(x_3, y[i + 3])
|
|
- 59 * self.func(x_2, y[i + 2])
|
|
+ 37 * self.func(x_1, y[i + 1])
|
|
- 9 * self.func(x_0, y[i])
|
|
)
|
|
x_0 = x_1
|
|
x_1 = x_2
|
|
x_2 = x_3
|
|
x_3 += self.step_size
|
|
|
|
return y
|
|
|
|
def step_5(self) -> np.ndarray:
|
|
"""
|
|
>>> def f(x,y):
|
|
... return x + y
|
|
>>> y = AdamsBashforth(
|
|
... f, [0, 0.2, 0.4, 0.6, 0.8], [0, 0.02140, 0.02140, 0.22211, 0.42536],
|
|
... 0.2, 1).step_5()
|
|
>>> y[-1]
|
|
0.05436839444444452
|
|
|
|
>>> AdamsBashforth(f, [0, 0.2, 0.4], [0, 0, 0.04], 0.2, 1).step_5()
|
|
Traceback (most recent call last):
|
|
...
|
|
ValueError: Insufficient initial points information.
|
|
"""
|
|
|
|
if len(self.x_initials) != 5 or len(self.y_initials) != 5:
|
|
raise ValueError("Insufficient initial points information.")
|
|
|
|
x_0, x_1, x_2, x_3, x_4 = self.x_initials[:5]
|
|
y_0, y_1, y_2, y_3, y_4 = self.y_initials[:5]
|
|
|
|
n = int((self.x_final - x_4) / self.step_size)
|
|
y = np.zeros(n + 6)
|
|
y[0] = y_0
|
|
y[1] = y_1
|
|
y[2] = y_2
|
|
y[3] = y_3
|
|
y[4] = y_4
|
|
|
|
for i in range(n + 1):
|
|
y[i + 5] = y[i + 4] + (self.step_size / 720) * (
|
|
1901 * self.func(x_4, y[i + 4])
|
|
- 2774 * self.func(x_3, y[i + 3])
|
|
- 2616 * self.func(x_2, y[i + 2])
|
|
- 1274 * self.func(x_1, y[i + 1])
|
|
+ 251 * self.func(x_0, y[i])
|
|
)
|
|
x_0 = x_1
|
|
x_1 = x_2
|
|
x_2 = x_3
|
|
x_3 = x_4
|
|
x_4 += self.step_size
|
|
|
|
return y
|
|
|
|
|
|
if __name__ == "__main__":
|
|
import doctest
|
|
|
|
doctest.testmod()
|