ODE solver methods (RK45, BDF) in SciPy - Step-by-Step Execution

Concept Flow - ODE solver methods (RK45, BDF)

Define ODE function

↓

Choose solver method: RK45 or BDF

↓

Set initial conditions and time span

↓

Run solver step-by-step

↓

Check solver status

↓

Collect solution points

↓

Plot or analyze results

The flow starts by defining the ODE, choosing a solver method (RK45 or BDF), setting initial conditions, running the solver stepwise, checking status, and collecting results.

Execution Sample

SciPy

from scipy.integrate import solve_ivp

def ode(t, y):
    return -2 * y

sol = solve_ivp(ode, [0, 5], [1], method='RK45')

This code solves the ODE dy/dt = -2y from t=0 to t=5 with initial y=1 using the RK45 method.

Execution Table

Step	t (time)	y (solution)	Method	Action	Status
1	0.0	[1.0]	RK45	Start integration	Running
2	0.1	[0.8187]	RK45	Compute next step	Running
3	0.2	[0.6703]	RK45	Compute next step	Running
4	0.3	[0.5488]	RK45	Compute next step	Running
5	0.4	[0.4493]	RK45	Compute next step	Running
6	0.5	[0.3679]	RK45	Compute next step	Running
7	5.0	[4.54e-05]	RK45	Final step reached	Success

💡 Integration ends when t reaches 5.0, the end of the time span.

Variable Tracker

Variable	Start	After Step 1	After Step 2	After Step 3	After Step 4	After Step 5	After Step 6	Final
t	0.0	0.1	0.2	0.3	0.4	0.5	5.0	5.0
y	[1.0]	[0.8187]	[0.6703]	[0.5488]	[0.4493]	[0.3679]	[4.54e-05]	[4.54e-05]
method	RK45	RK45	RK45	RK45	RK45	RK45	RK45	RK45

Key Moments - 2 Insights

Why does the solver take smaller steps at the start and then a big jump to the final time?

The solver adapts step size to keep accuracy. Early steps are small to capture rapid changes; later it can jump to the final time if error is low. See execution_table rows 2-6 for small steps and row 7 for the big jump.

What's the difference between RK45 and BDF methods in solving ODEs?

RK45 is an explicit method good for smooth problems; BDF is implicit and better for stiff problems. Both solve step-by-step but handle stability differently. This example uses RK45, but switching method='BDF' changes solver behavior.

Visual Quiz - 3 Questions

Look at the execution_table, what is the approximate value of y at step 4?

A [0.5488]

B [0.6703]

C [0.4493]

D [0.3679]

💡 Hint

Check the 'y (solution)' column at step 4 in the execution_table.

At which step does the solver reach the final time t=5.0?

A Step 5

B Step 7

C Step 6

D Step 4

💡 Hint

Look at the 't (time)' column in execution_table for the value 5.0.

If we change the method to 'BDF', what would most likely happen to the step sizes?

A They become larger and less adaptive

B They stay exactly the same as RK45

C They become smaller and more stable for stiff problems

D The solver stops immediately

💡 Hint

Refer to key_moments about method differences and solver stability.

Concept Snapshot

Solve ODEs with solve_ivp:
- Define function ode(t, y)
- Choose method='RK45' (explicit) or 'BDF' (implicit)
- Set time span and initial y
- Run solver stepwise
- Adaptive step sizes for accuracy
- RK45 good for smooth, BDF for stiff problems

Full Transcript

This visual execution shows how to solve an ordinary differential equation (ODE) using scipy's solve_ivp function with the RK45 method. We start by defining the ODE function dy/dt = -2y. The solver begins at time t=0 with initial value y=1. It takes small steps, computing y values at each step, adapting step size to keep accuracy. The solution y decreases exponentially as expected. The solver finishes when it reaches the final time t=5. The RK45 method is explicit and good for smooth problems. The BDF method, not shown here, is implicit and better for stiff problems. Step sizes and solver behavior differ between methods. This step-by-step trace helps beginners see how the solver progresses through time and updates solution values.

Practice

1. Which scipy ODE solver method is best suited for stiff problems?

easy

A. BDF

B. RK45

C. Euler

D. Midpoint

Solution

1. Step 1: Understand stiff problems

   Stiff problems require methods that handle rapid changes without instability.

2. Step 2: Match solver to problem type

   BDF (Backward Differentiation Formula) is designed for stiff problems, unlike RK45 which is for general cases.

3. Final Answer:

   BDF -> Option A

4. Quick Check:

   Stiff problems = BDF [OK]

Hint: BDF is for stiff problems, RK45 for general ones [OK]

Common Mistakes:

• Confusing RK45 as best for stiff problems
• Thinking Euler method is in solve_ivp
• Assuming Midpoint is a scipy method

2. Which of the following is the correct way to specify the RK45 method in solve_ivp?

easy

A. method='RungeKutta45'

B. method='rk45'

C. method='RK_45'

D. method='RK45'

Solution

1. Step 1: Check method parameter syntax

   The solve_ivp function expects method names as strings exactly matching documented names.

2. Step 2: Verify correct spelling and case

   The correct method name is 'RK45' with uppercase letters and no underscores or spaces.

3. Final Answer:

   method='RK45' -> Option D

4. Quick Check:

   Exact method name = 'RK45' [OK]

Hint: Use exact method names with correct case in solve_ivp [OK]

Common Mistakes:

• Using lowercase 'rk45' instead of 'RK45'
• Adding underscores like 'RK_45'
• Using descriptive names not accepted by solve_ivp

3. What will be the output of this code snippet?

from scipy.integrate import solve_ivp
import numpy as np

def fun(t, y):
    return -2 * y

sol = solve_ivp(fun, [0, 2], [1], method='RK45', t_eval=[0, 1, 2])
print(sol.y[0])

medium

A. [1. 0.13533528 0.01831564]

B. [1. 0.5 0.25]

C. [1. 2. 4.]

D. [0. 0. 0.]

Solution

1. Step 1: Understand the ODE and solution

   The ODE is dy/dt = -2y with initial y(0)=1, which has exact solution y(t) = exp(-2t).

2. Step 2: Calculate values at t=0,1,2

   y(0)=1, y(1)=exp(-2) ≈ 0.1353, y(2)=exp(-4) ≈ 0.0183.

3. Final Answer:

   [1. 0.13533528 0.01831564] -> Option A

4. Quick Check:

   exp(-2t) at t=0,1,2 = [1. 0.13533528 0.01831564] [OK]

Hint: Use exact solution exp(-2t) to check output array [OK]

Common Mistakes:

• Confusing exponential decay with linear decay
• Misreading t_eval points
• Assuming RK45 returns integers

4. Identify the error in this code using BDF method:

from scipy.integrate import solve_ivp

def f(t, y):
    return y - t**2 + 1

sol = solve_ivp(f, [0, 2], 0.5, method='BDF', t_eval=[0, 1, 2])
print(sol.y)

medium

A. Function f must take only one argument

B. Initial condition y0 should be a list or array, not a float

C. Method 'BDF' is not supported in solve_ivp

D. t_eval must be a numpy array, not a list

Solution

1. Step 1: Check initial condition format

   The initial condition y0 must be an array-like object, even if single value, e.g., [0.5].

2. Step 2: Verify given y0 type

   Here y0 is given as 0.5 (float), which is not array-like and causes error.

3. Step 3: Confirm other parameters

   Function f has correct signature, method 'BDF' is valid, t_eval can be list or array.

4. Final Answer:

   Initial condition y0 should be a list or array, not a float -> Option B

5. Quick Check:

   y0 must be array-like [OK]

Hint: Always pass y0 as list or array, not single float [OK]

Common Mistakes:

• Passing y0 as float instead of list
• Thinking BDF is unsupported
• Believing t_eval must be numpy array

5. You want to solve a stiff ODE system and compare solutions from RK45 and BDF methods. Which approach correctly compares their results at specific times?

hard

A. Use only RK45 since BDF is slow and ignore BDF results

B. Run RK45 without t_eval and BDF with t_eval, then compare sol.t arrays

C. Use solve_ivp with t_eval for both methods, then compare sol.y arrays at those times

D. Compare the initial conditions only, since methods differ internally

Solution

1. Step 1: Understand comparison requirements

   To compare solutions, evaluate both methods at the same time points using t_eval.

2. Step 2: Use solve_ivp with t_eval for both

   Passing t_eval ensures outputs have matching time points, so sol.y arrays can be directly compared.

3. Step 3: Reject incorrect approaches

   Not using t_eval leads to different time points; ignoring BDF or comparing only initial conditions misses solution differences.

4. Final Answer:

   Use solve_ivp with t_eval for both methods, then compare sol.y arrays at those times -> Option C

5. Quick Check:

   Same t_eval for both = valid comparison [OK]

Hint: Use t_eval to get matching time points for comparison [OK]

Common Mistakes:

• Not using t_eval for both methods
• Comparing sol.t arrays instead of sol.y
• Ignoring BDF results due to speed concerns