25.12.2. Ill-Conditioned System

An ill-conditioned system is highly sensitive to small changes in coefficients or data, leading to large errors in the solution.

An ill-conditioned system refers to a scenario where slight variations in input data or the coefficients of the equations can cause significant fluctuations in the output solution. This vulnerability occurs when the matrix representing the system has a high condition number, indicating that small errors in data can exponentially amplify.

Imagine you are balancing a pencil on your finger. If you jiggle your finger even slightly, the pencil falls off easily: this is similar to how an ill-conditioned system reacts to minor changes. In contrast, a well-balanced object, like a heavy book on a flat surface, requires a far larger disturbance to cause it to topple over.

This is measured by the condition number of matrix A: κ(A)=∥A∥⋅∥A−1∥.

• κ(A)≫1: Ill-conditioned
• κ(A)≈1: Well-conditioned

The condition number (κ) is a quantitative measure that indicates how sensitive a matrix is to perturbations. It is calculated as the product of the norm of the matrix and the norm of its inverse. If κ is much greater than 1, the matrix is deemed ill-conditioned, signifying a high sensitivity to changes in input. Conversely, a condition number close to 1 suggests a well-conditioned matrix, implying stability against minor changes.

Think of a bridge's design behavior under varying loads. A well-designed bridge (well-conditioned system) can handle small weight variations without collapsing. In contrast, an ill-designed bridge (ill-conditioned system) might buckle under slightly varying loads, indicating poor structural integrity.

Such systems are common in:

• Statically indeterminate structures with close-to-parallel constraints.
• Long-span bridge modeling.

Ill-conditioned systems often appear in specific engineering scenarios, such as those involving statically indeterminate structures. These situations typically occur when the constraints are nearly parallel, making the system highly sensitive to any changes. For instance, long-span bridges are another example of conditions that often lead to ill-conditioning due to their intricate balance and sensitivity to loads and supports.

Consider trying to balance multiple long, thin sticks on top of one another. If the sticks are too similar in angle or position, a tiny disturbance can cause a collapse. Similarly, if engineers make a small error when modeling a long bridge, this can dramatically affect its stability, much like the precarious tower of sticks.

Remedies:

• Use pivoting in Gaussian elimination.
• Improve numerical precision.
• Reformulate the model if necessary.

To manage the issues arising from ill-conditioned systems, several approaches can be utilized. Pivoting during Gaussian elimination helps to enhance the numerical stability of the system. Additionally, improving numerical precision—such as using higher-accuracy calculations—can help mitigate errors. If these solutions do not suffice, reformulating the model to simplify system equations and relationships may also be necessary.

When making a complex dish, if one ingredient is missing or measured incorrectly, the dish might fail. Chefs might pivot and add a substitute (pivoting), measure the ingredients more carefully (improving numerical precision), or change the recipe altogether (reformulate the model) to produce a better outcome.

--