Unit 4: Numerical Solution of Ordinary Differential Equations - Subjective Questions

Initial Value Problem (IVP): An initial value problem consists of a first-order differential equation together with an initial condition:

The objective is to determine the function satisfying both the differential equation and the specified initial condition.

Existence condition: If is continuous in a rectangular region containing , then at least one solution exists in some interval containing .

Uniqueness condition: If both and its partial derivative are continuous in the region, then the solution is unique in some interval containing .

These conditions are important because numerical methods approximate a solution that should exist and be uniquely determined by the initial data.

Numerical solution procedure:

• Divide the interval of interest into equally spaced points using step size :
  
• Start with the known initial value .
• Use a numerical formula to calculate an approximation from and .
• Repeat the process until the required value of is reached.

For a general one-step method, the solution is written as

where is an approximation to the derivative over the interval .

The main sources of error are truncation error, round-off error, and errors propagated from earlier steps. A smaller step size generally improves accuracy, but it increases computational effort.

Using Taylor's expansion of about ,

Since , neglecting terms containing and higher powers gives

This is called Euler's method.

Algorithm:

• Evaluate the slope .
• Multiply the slope by the step size .
• Add the result to .

Thus,

The local truncation error is of order , while the global truncation error is of order . Therefore, Euler's method is a first-order method.

For Euler's method,

Here, , , , and .

First step:

Thus, and .

Second step:

Therefore,

The result is an approximation because Euler's method neglects higher-order derivative terms.

The Modified Euler method improves Euler's method by using the average of the slopes at the beginning and end of the interval.

For the differential equation

first obtain a predicted value using Euler's formula:

Next, evaluate the slope at the predicted point and use the average slope:

The steps are:

• Calculate the initial slope .
• Predict using Euler's method.
• Calculate the terminal slope .
• Correct the value using the average slope.

The local truncation error is and the global truncation error is , so it is a second-order method.

In Modified Euler's method,

and

The improved accuracy is obtained by representing the solution curve using an average slope instead of assuming that the initial slope remains constant throughout the interval.

Runge-Kutta methods approximate the weighted average slope over an interval without explicitly calculating higher derivatives. They are one-step methods because each new value is calculated from the current point only.

For

and step size , the fourth-order Runge-Kutta method, or RK4, is defined by

The next approximation is

The values , , , and represent slopes at the beginning, two midpoint estimates, and the end of the interval. RK4 has global truncation error .

The RK4 formula is obtained by evaluating the derivative at four carefully selected points in the interval .

The four slopes are

The weighted average slope is

Therefore, the approximate solution at the next point is

or

The weights , , , and provide fourth-order accuracy. The local error is and the global error is .

Second-order Runge-Kutta method:

A common second-order form uses two slope evaluations:

Its global truncation error is .

Fourth-order Runge-Kutta method:

RK4 uses four slope evaluations:

Its global truncation error is .

Differences:

• RK2 requires fewer function evaluations and is computationally cheaper.
• RK4 requires more computations but gives substantially higher accuracy for the same step size.
• RK2 is suitable when moderate accuracy is sufficient.
• RK4 is preferred in robotics simulations when reliable trajectory and state estimates are required.
• Neither method requires values from earlier steps beyond the current point.

Multi-step methods calculate using values of the solution and derivative at several previous points. A general linear multi-step formula can be written as

where .

Advantages:

• They can achieve high accuracy with fewer function evaluations per step.
• They reuse previously computed derivative values.
• They are efficient for long integrations.

Disadvantages:

• They require several starting values.
• Starting values must usually be generated by a one-step method such as RK4.
• Errors may propagate through successive steps.
• Stability analysis is more complicated.
• They are less convenient when the step size changes frequently.

Examples include Adams-Bashforth, Adams-Moulton, Milne, and Hamming methods.

Consider

Integrating from to gives

Approximate by a cubic interpolation polynomial through the four known values , , , and . Integrating this interpolation polynomial over one step gives

This is the four-step Adams-Bashforth formula.

It is explicit because does not appear on the right-hand side. The method has fourth-order accuracy, with global truncation error . Four starting values are needed, and these can be calculated using RK4.

The Milne method is a four-step predictor-corrector method. It uses four previous solution and derivative values to estimate the next value.

Let

The Milne predictor is

After calculating the predicted value, evaluate

The Milne corrector is

The corrected value may be substituted again into the corrector formula if greater accuracy is required.

The method requires four initial values, usually generated by RK4. The predictor is explicit, while the corrector is implicit in form because it involves the derivative at the new point.

The Milne method needs the four values , , , and before it can calculate . Since only is given by the initial value problem, the remaining values are generated using a one-step method, commonly RK4.

For each subsequent step:

• Compute the known derivative values .
• Use the predictor formula:
  
• Evaluate the derivative at the predicted point:
  
• Apply the corrector formula:
  
• Replace the predicted value by the corrected value.

The corrected value can be iterated until successive corrections differ by less than a selected tolerance.

Single-step methods are convenient and robust, especially when the solution is started or the step size changes. Multi-step methods are computationally efficient after sufficient previous values have been generated.

Local truncation error: This is the error introduced in one step when the exact solution is substituted into the numerical formula. It measures the error caused by approximating the differential equation over a single interval.

For a method of order , the local truncation error is generally

Global truncation error: This is the accumulated error after many steps of computation. For a stable method of order , it is generally

For example:

• Euler's method has local error and global error .
• Modified Euler's method has local error and global error .
• RK4 has local error and global error .

These errors help determine the step size and method required to achieve a specified accuracy.

Consistency: A numerical method is consistent if its local truncation error tends to zero as . The method must correctly represent the original differential equation in the limiting case.

Stability: A method is stable if small errors in the initial data, round-off, or previous numerical values do not grow without bound during computation. Stability is particularly important for systems whose solutions contain rapidly decaying components.

Convergence: A numerical method is convergent if the numerical solution approaches the exact solution as :

For many linear initial value problems, consistency together with stability implies convergence. A method may be accurate in theory but unsuitable in practice if it is unstable for the chosen step size.

For the test equation

Euler's method gives

Thus, the amplification factor is

For numerical stability, the magnitude of the amplification factor must satisfy

If is real and negative, this becomes

which gives

or, equivalently,

Therefore, Euler's method requires a sufficiently small step size for stable computation of decaying systems. This issue is important in robotics because dynamic models may contain fast and slow modes simultaneously.

Robot motion is commonly described by differential equations involving position, velocity, acceleration, forces, and torques. For a robot state vector , the dynamics can be expressed as

where represents control inputs.

For a simple joint with angle , angular velocity , inertia , and applied torque , the equations may be written as

These equations are often nonlinear and do not have closed-form solutions. Numerical methods are therefore used to:

• Simulate robot trajectories.
• Predict position and velocity.
• Design and test controllers.
• Estimate future states in real time.
• Study stability and dynamic performance.

Methods such as RK4 provide accurate offline simulation, while simpler methods may be preferred for real-time embedded computation.

Let be generalized coordinates. Define the Lagrangian as

where is kinetic energy and is potential energy.

Using the principle of virtual work and generalized forces , the equation of motion for the th coordinate is

for .

If all applied forces are conservative and included in , then , giving

For a robot manipulator, the generalized coordinates are usually joint variables. The resulting equations are second-order ordinary differential equations and can be converted into a first-order state-space form before applying Euler, Runge-Kutta, or multi-step methods.

Consider a pendulum of mass and length , with angular displacement from the vertically downward direction.

The velocity of the mass is , so the kinetic energy is

Taking the lowest position as the zero potential energy reference, the potential energy is

Therefore, the Lagrangian is

Apply Lagrange's equation:

Now,

and

Hence,

or