4. Nonlinear Algebraic Equations

Introduction and Context

Many engineering models reduce, after discretization or physical simplification, to the problem of finding an unknown vector x that satisfies

\[ \mathbf{F}(\mathbf{x})=\mathbf{0}. \]

For a single unknown, this becomes \(f(x)=0\). Such equations are called nonlinear when the unknown does not occur only to the first power and is not restricted to linear combinations of known functions. Polynomial equations of degree greater than one, trigonometric equations, exponential relations, constitutive laws, contact conditions, and equilibrium equations with temperature-dependent properties are common examples.

Nonlinear algebraic equations arise throughout mechanical engineering. A pressurized vessel may require a wall thickness satisfying a nonlinear stress relation; a fluid-flow calculation may involve a friction factor defined implicitly; a heat-transfer model may contain radiation terms proportional to \(T^4\); and a mechanism or finite-element model may require equilibrium under geometrically nonlinear deformation. In each case, the central computational task is not merely to obtain a numerical value, but to determine whether the value is physically meaningful, whether the algorithm has converged reliably, and how sensitive the result is to data and initial guesses.

Unlike linear systems, nonlinear problems may possess several solutions, no solution, or solutions that are difficult to locate. Convergence can depend strongly on the starting point. A method may converge rapidly for one initial estimate and fail completely for another. Consequently, nonlinear solution is both a mathematical and an engineering decision-making activity.

Learning Objectives

After studying this chapter, the reader should be able to:

1. Formulate scalar and coupled engineering problems as nonlinear algebraic equations.

2. Explain existence, uniqueness, conditioning, residuals, and convergence.

3. Apply bisection, regula falsi, fixed-point iteration, Newton’s method, and quasi-Newton methods.

4. Select appropriate methods according to robustness, derivative availability, cost, and problem structure.

5. Extend scalar ideas to coupled nonlinear systems.

6. Use continuation and damping to improve robustness.

7. Assess numerical results using residuals, step sizes, physical constraints, and sensitivity.

8. Recognize implementation errors, failure modes, and professional responsibilities in nonlinear computation.

4.1 The Mathematical Setting

A scalar nonlinear equation is written as

\[ f(x)=0. \]

A solution \(x^\) is a root of \(f\). The residual at an approximation \(x_k\) is \(r_k=f(x_k)\). A small residual indicates that the equation is nearly satisfied, but does not by itself guarantee that the error \(|x_k-x^|\) is small. If the slope near the root is very small, a small residual can correspond to a relatively large error.

For a system of \(n\) equations,

\[ \mathbf{F}(\mathbf{x})=

\begin{bmatrix}f_1(\mathbf{x})\\ \vdots \\ f_n(\mathbf{x})\end{bmatrix}=\mathbf{0}, \qquad

\mathbf{x}\in\mathbb{R}^n. \]

The Jacobian matrix is

\[ \mathbf{J}(\mathbf{x})=\frac{\partial \mathbf{F}}{\partial \mathbf{x}}. \]

Its element in row \(i\), column \(j\) is \(\partial f_i/\partial x_j\). The Jacobian is the multidimensional analogue of the derivative and controls both local sensitivity and Newton-type corrections.

Existence, uniqueness, and conditioning

For a continuous scalar function, if \(f(a)f(b)<0\), the intermediate value theorem guarantees at least one root in \([a,b]\). This is an existence result, not a uniqueness result. Several roots may lie within the interval. If \(f\) is strictly monotonic there, the root is unique.

A root is simple if \(f'(x^)\neq 0\). A multiple root has \(f'(x^)=0\), and generally causes slower convergence and greater sensitivity. The local condition of a scalar root is related to \(1/|f'(x^*)|\): a small derivative means that perturbations in the equation or data can produce larger changes in the root.

For systems, nonsingularity of \(\mathbf{J}(\mathbf{x}^*)\) is the local condition for an isolated, well-behaved solution. A nearly singular Jacobian indicates ill-conditioning. In such circumstances, even a numerically converged answer may be highly sensitive to measurement errors, material-property uncertainty, or rounding.

Convergence language

An iteration \(x_{k+1}=\Phi(x_k)\) converges to \(x^\) if \(x_k\to x^\). It has order \(p\) if, approximately,

\[ |e_{k+1}|\approx C|e_k|^p, \qquad e_k=x_k-x^*. \]

Linear convergence corresponds to \(p=1\), quadratic convergence to \(p=2\), and superlinear convergence to a rate faster than linear but not necessarily quadratic. The attractive speed of Newton’s method is local: it is normally quadratic near a simple root, but only when the initial estimate is sufficiently close and the derivative is reliable.

4.2 Bracketing Methods: Reliability Before Speed

Bracketing methods maintain an interval known to contain a root. They are valuable when a guaranteed, predictable method is more important than rapid local convergence.

Bisection

Suppose \(f(a)f(b)<0\). The midpoint is

\[ c=\frac{a+b}{2}. \]

If \(f(a)f(c)<0\), replace \(b\) with \(c\); otherwise replace \(a\) with \(c\). The process continues until the interval width or a residual criterion is sufficiently small.

After \(k\) bisections, the interval width is

\[ \frac{b-a}{2^k}. \]

Thus, bisection has guaranteed linear convergence for continuous functions with a sign-changing bracket. Its principal strengths are robustness, simplicity, and an explicit error bound. Its weaknesses are slow convergence and inability to exploit derivative information.

A sound stopping rule may require both

\[ |b-a|\leq \varepsilon_x(1+|c|) \quad\text{and}\quad |f(c)|\leq\varepsilon_f. \]

Using only a residual can be unsafe near a flat root; using only an interval width can be unsafe when function evaluation is poorly scaled.

Regula falsi and safeguarded interpolation

The false-position method uses the secant line through \((a,f(a))\) and \((b,f(b))\):

\[ c=b-f(b)\frac{b-a}{f(b)-f(a)}. \]

The sign-changing bracket is retained. Regula falsi can be faster than bisection, but one endpoint may remain fixed for many iterations, producing slow progress. Variants such as the Illinois method modify function values to prevent endpoint stagnation.

In practical software, a hybrid method often combines interpolation with bisection. An interpolation step is accepted only if it remains inside the bracket and appears numerically safe; otherwise, the algorithm takes a bisection step. This combines speed with a fallback guarantee.

When bracketing is the right choice

Bracketing is especially appropriate when the function is continuous, a physical interval is available, derivatives are unavailable or discontinuous, and failure is costly. Examples include finding a critical temperature within a known operating range or solving for a friction factor constrained to be positive. A bracket should be based on physical reasoning rather than arbitrary large numbers, because overflow, invalid property correlations, or nonphysical states may occur outside the operating range.

4.3 Fixed-Point Iteration

A nonlinear equation can sometimes be rearranged as

\[ x=g(x), \]

leading to the iteration

\[ x_{k+1}=g(x_k). \]

The iteration converges locally if, in a neighborhood of the fixed point,

\[ |g'(x)|<1. \]

A sufficient condition on an interval is \(|g'(x)|\leq L<1\), together with \(g(x)\) mapping the interval into itself. Then the fixed point is unique in that interval and the iteration is contractive.

The rearrangement is not a harmless algebraic detail. For example, an equation may have two equivalent forms, one with \(|g'|<1\) and another with \(|g'|>1\). The first converges while the second diverges or oscillates. This makes fixed-point iteration a method-design problem.

If the iteration oscillates, relaxation may help:

\[ x_{k+1}=(1-\omega)x_k+\omega g(x_k), \qquad 0<\omega<1. \]

Under-relaxation reduces step size and may stabilize a calculation, although it also slows convergence. Over-relaxation can accelerate some problems but is more likely to cause instability. For systems, the same idea is written \(\mathbf{x}_{k+1}=\mathbf{G}(\mathbf{x}_k)\), with convergence related to the spectral radius of \(\partial\mathbf{G}/\partial\mathbf{x}\).

Fixed-point methods are attractive when the physical model naturally supplies an update equation, when derivatives are difficult to calculate, or when a low-cost iteration is needed as an initial solver. Their convergence should never be assumed merely because the rearrangement is algebraically correct.

4.4 Newton’s Method and Its Safeguards

For a scalar equation, Newton’s method is

\[ x_{k+1}=x_k-\frac{f(x_k)}{f'(x_k)}. \]

Geometrically, the tangent at \(x_k\) is extended to its intersection with the axis. Near a simple root, Newton’s method is quadratically convergent. The method is computationally efficient because each step uses local slope information rather than retaining a bracket.

However, Newton’s method is not globally reliable. Failure may result when the derivative is zero or very small, the initial guess is distant from the root, the function has a sharp curvature, or the iteration enters a region where the model is undefined. It may jump outside a physically permissible domain or converge to a different root.

A practical implementation uses safeguards. A step may be limited according to

\[ x_{k+1}=x_k+\alpha_k\Delta x_k, \qquad \Delta x_k=-f(x_k)/f'(x_k), \quad 0<\alpha_k\leq1. \]

The factor \(\alpha_k\) is selected by line search so that the residual decreases and the new point remains admissible. If a bracket is available, the Newton point can be accepted only when it lies inside the bracket; otherwise, bisection is used. This is the basis of safeguarded Newton and several hybrid root-finding algorithms.

Finite-difference derivatives are often used when analytic derivatives are unavailable:

\[ f'(x)\approx\frac{f(x+h)-f(x-h)}{2h}. \]

The step \(h\) must balance truncation error against round-off error. Derivative verification, automatic differentiation, or complex-step differentiation can be valuable in high-accuracy implementations.

4.5 Quasi-Newton Methods for Coupled Systems

Newton’s method for a system solves

\[ \mathbf{J}(\mathbf{x}_k)\Delta\mathbf{x}_k=-\mathbf{F}(\mathbf{x}_k), \qquad \mathbf{x}_{k+1}=\mathbf{x}_k+\Delta\mathbf{x}_k. \]

The Jacobian should not normally be inverted explicitly. Instead, the linear system is solved using LU, QR, or an appropriate sparse factorization. Explicit inversion is less stable and usually more expensive.

For large or repeatedly solved problems, forming and factoring the exact Jacobian at every iteration may dominate computational cost. Quasi-Newton methods update an approximation \(\mathbf{B}_k\) to the Jacobian or an approximation to its inverse. The secant condition requires

\[ \mathbf{B}_{k+1}\mathbf{s}_k=\mathbf{y}_k, \quad \mathbf{s}_k=\mathbf{x}_{k+1}-\mathbf{x}_k, \quad \mathbf{y}_k=\mathbf{F}(\mathbf{x}_{k+1})-\mathbf{F}(\mathbf{x}_k). \]

The Broyden family is widely used because it updates the Jacobian approximation from function evaluations without requiring all derivatives. It often provides superlinear convergence near a solution and is useful when function evaluations are expensive or derivatives are difficult to obtain.

Quasi-Newton methods can nevertheless degrade when the Jacobian approximation becomes inaccurate, variables have very different scales, or the problem is poorly conditioned. Periodic Jacobian recomputation, scaling, damping, and restart strategies are common remedies.

4.6 Continuation: Tracking Solutions Rather Than Guessing Them

Some nonlinear problems are difficult because the desired solution is not close to a convenient initial guess. Continuation introduces a parameter \(\lambda\) and solves a sequence of related problems:

\[ \mathbf{F}(\mathbf{x},\lambda)=\mathbf{0}, \qquad 0\leq\lambda\leq1. \]

At \(\lambda=0\), the problem is chosen to be easy. The solution at one parameter value becomes the initial guess for the next. For example, a load can be increased gradually from zero to its target value, or a nonlinear material law can be introduced progressively.

Simple parameter continuation may fail at a limit point, where the solution curve turns and \(d\mathbf{x}/d\lambda\) becomes unbounded. Pseudo-arclength continuation overcomes this difficulty by treating both \(\mathbf{x}\) and \(\lambda\) as unknowns and imposing a geometric path constraint. It is particularly important in nonlinear structural mechanics, buckling, multiphysics, and bifurcation analysis.

Continuation is more than a numerical trick. It provides information about the full response path, including multiple branches, turning points, and loss of stability. The computed equilibrium branch, however, should not automatically be interpreted as a dynamically stable operating state; stability requires a separate physical analysis.

4.7 A Practical Solution Workflow

A reliable nonlinear computation begins before an algorithm is selected. First formulate the residual so that every term has consistent units and the intended physical constraints are explicit. Identify possible roots, admissible ranges, singularities, and discontinuities. Plotting or sampling the residual over a physically meaningful range often reveals whether a bracket exists and whether multiple roots are likely.

Next choose an algorithm. A useful decision pattern is:

• use bisection or a safeguarded hybrid when a bracket and guaranteed progress are essential;
• use fixed-point iteration when a contractive physical update is available;
• use Newton’s method when derivatives are reliable and a good initial estimate exists;
• use quasi-Newton methods when systems are coupled and derivative construction is costly;
• use continuation when the target problem is difficult to reach directly.

Scale variables and equations so that their typical magnitudes are comparable. For a system, a dimensionless variable may be formed as \(\hat{x}_i=x_i/s_i\), where \(s_i\) is a characteristic scale. Similarly, residual scaling prevents one equation with large numerical units from dominating the stopping test.

A robust termination test should consider at least the residual norm and the step norm:

\[ \|\mathbf{F}(\mathbf{x}_k)\|\leq\varepsilon_F, \qquad

\|\Delta\mathbf{x}_k\|\leq\varepsilon_x(1+\|\mathbf{x}_k\|). \]

Iteration limits, stagnation detection, invalid-state checks, and logging should also be implemented. After convergence, substitute the computed solution into the original, unscaled equations, check all constraints, and perturb initial guesses or parameters to assess sensitivity.

4.8 Illustrative Engineering Examples

Scalar example: a radiation-convection balance

Suppose a surface temperature \(T\) is determined by

\[ f(T)=\varepsilon\sigma A(T^4-T_\infty^4)+hA(T-T_\infty)-Q=0. \]

Here \(\varepsilon\) is emissivity, \(\sigma\) is the Stefan–Boltzmann constant, \(A\) is area, \(h\) is a convection coefficient, and \(Q\) is supplied heat. The physically admissible domain is typically \(T>0\). Since the radiation term grows rapidly with temperature, a bracket can often be obtained by evaluating the balance at a lower and upper physically credible temperature.

Bisection gives a guaranteed answer if the signs differ. Newton’s method uses

\[ f'(T)=4\varepsilon\sigma AT^3+hA. \]

Because this derivative is positive for positive parameters, the balance is monotonic and the root is unique in the admissible domain. A sensible workflow is to bracket the solution, use a safeguarded Newton step, and verify the final energy residual rather than relying only on a temperature-change criterion.

Coupled example: nonlinear spring equilibrium

Consider two displacement coordinates \(u_1,u_2\) governed by

\[ F_1=k_1u_1+c(u_1-u_2)^3-P_1=0, \]

\[ F_2=k_2u_2-c(u_1-u_2)^3-P_2=0. \]

The Jacobian is

\[ \mathbf{J}=\begin{bmatrix}

k_1+3c(u_1-u_2)^2 & -3c(u_1-u_2)^2\\

-3c(u_1-u_2)^2 & k_2+3c(u_1-u_2)^2

\end{bmatrix}. \]

At each Newton iteration, the linear system \(\mathbf{J}\Delta\mathbf{u}=-\mathbf{F}\) is solved. If the loads are large or the initial displacement is poor, load continuation can be used: replace \(P_i\) by \(\lambda P_i\), increase \(\lambda\) in small increments, and use each converged displacement as the next initial guess.

4.9 Applied Case Study: Nonlinear Flow Resistance

Background

An illustrative piping calculation requires the Darcy friction factor \(f_D\) for turbulent flow in a rough circular pipe. The friction factor is defined implicitly by the Colebrook–White relation

\[ \frac{1}{\sqrt{f_D}}=-2\log_{10}\left(\frac{\epsilon/D}{3.7}+\frac{2.51}{Re\sqrt{f_D}}\right), \]

where \(Re\) is Reynolds number and \(\epsilon/D\) is relative roughness.

Situation and problem

The design calculation needs a reliable friction factor for use in pressure-drop and pump-sizing estimates. The equation is nonlinear, the unknown must be positive, and the logarithm requires a positive argument. A careless Newton iteration can produce a negative or invalid trial value.

Analysis

Define

\[ g(f_D)=\frac{1}{\sqrt{f_D}}+2\log_{10}\left(\frac{\epsilon/D}{3.7}+\frac{2.51}{Re\sqrt{f_D}}\right). \]

For typical turbulent conditions, engineering correlations provide a physically reasonable interval, for example a small positive range bounded by smooth- and rough-pipe estimates. The function can be sampled over that interval to verify a sign change and to detect unexpected multiple roots. The variable may also be transformed using \(y=1/\sqrt{f_D}\), which improves numerical behavior and guarantees positivity when \(f_D=1/y^2\).

Approach and solution

A robust implementation first constructs a positive bracket and applies a safeguarded secant or bisection method. Newton’s method may then be used within the bracket to accelerate convergence. The stopping test includes the Colebrook residual, relative change in friction factor, and validity of the logarithm argument. The result is substituted back into the original relation, not merely into a rearranged computational form.

Outcome

The method produces a friction factor satisfying the implicit equation while preserving the physical domain. The pressure-drop calculation can then proceed, but its uncertainty should include uncertainty in roughness, Reynolds number, and the validity of the chosen pipe-correlation regime.

Learning points

The case demonstrates that a good nonlinear algorithm must respect physical constraints, not merely reduce an algebraic residual. Bracketing supplies robustness, transformation can enforce admissibility, and verification against the original equation protects against coding or rearrangement errors.

4.10 Challenges, Risks, and Common Mistakes

A frequent mistake is to report convergence because successive iterates are close without checking the residual. Another is to use an unscaled norm for a system whose variables have units and widely different magnitudes. Numerical derivatives may be inaccurate because of inappropriate finite-difference steps. Explicit Jacobian inversion, unprotected Newton steps, and failure to check domain restrictions are also common sources of unreliable results.

Multiple roots require special care. A sign-changing bracket may not exist for an even-multiplicity root, so scanning only for sign changes can miss physically important solutions. Conversely, finding one root does not establish uniqueness. Discontinuous empirical correlations, piecewise material laws, and hysteresis can invalidate methods that assume smoothness.

The solver may converge to a mathematically valid but physically irrelevant solution. Engineers must therefore impose positivity, conservation, operating limits, and constitutive validity independently of the numerical algorithm. If the Jacobian is singular, the issue may be a genuine bifurcation, redundant equations, poor scaling, or an implementation error; it should not be concealed by simply increasing the iteration limit.

4.11 Professional and Computational Considerations

Nonlinear results can influence safety margins, equipment sizing, thermal limits, and control decisions. Reproducible computation requires documenting equations, units, parameter values, initial guesses, tolerances, solver versions, and convergence histories. Important calculations should be independently checked through an alternative formulation, limiting-case analysis, or a second numerical method.

Automatic differentiation, sparse linear algebra, parallel function evaluation, and scientific computing libraries have made large nonlinear solves more practical. Machine-learning surrogates may reduce the cost of repeated function evaluations, but a surrogate can introduce bias, violate conservation, or behave poorly outside its training domain. It should therefore be validated against the governing model and used with domain monitoring.

In safety-critical work, an apparently converged numerical answer is not sufficient evidence of correctness. Verification asks whether the equations were implemented correctly; validation asks whether the equations represent the physical system adequately. Both are required.

4.12 Current Directions and Best Practice

Modern nonlinear solvers increasingly combine globalization techniques, such as trust regions and line searches, with fast local models. Trust-region methods restrict the step to a region where the local approximation is considered credible, often improving behavior when Newton steps are too aggressive. Large-scale applications use matrix-free Newton–Krylov methods, in which Jacobian-vector products are computed without explicitly storing the Jacobian.

Best practice can be summarized as follows:

1. Formulate a well-scaled residual and state the physical domain.

2. Establish existence, possible multiplicity, and useful bounds where possible.

3. Prefer safeguarded methods when failure is costly.

4. Use accurate derivatives or verify numerical derivatives.

5. Monitor residuals, steps, condition indicators, and admissibility.

6. Use continuation, damping, or trust regions for difficult problems.

7. Verify the final result in the original equations and assess sensitivity.

8. Record computational choices so the result is reproducible.

Chapter Summary

Nonlinear algebraic equations model equilibrium, transport, constitutive behavior, and many other engineering phenomena. Scalar equations may be solved with robust bracketing methods, fixed-point iteration, or rapidly convergent Newton iterations. Bracketing guarantees containment under continuity and a sign change; fixed-point convergence depends on contraction; Newton’s method is locally quadratic but requires safeguards away from the root. Coupled systems require Jacobians and linear solves, while quasi-Newton methods reduce derivative cost through updates. Continuation transforms difficult target problems into sequences of manageable problems and can reveal entire solution branches. Reliable practice combines mathematical convergence with scaling, physical admissibility, residual verification, sensitivity assessment, and documented computation.

Key Terms

Residual: The value of the governing equation at an approximate solution.

Root: A value satisfying a scalar nonlinear equation exactly in mathematical terms.

Jacobian: Matrix of first partial derivatives of a vector residual function.

Bracketing: Maintaining an interval known to contain a root.

Bisection: A bracketing method that repeatedly halves an interval.

Regula falsi: A bracketing method based on a secant-line intersection.

Fixed point: A value satisfying \(x=g(x)\).

Contraction: A mapping that reduces distances between points.

Newton’s method: An iteration based on a local tangent or Jacobian linearization.

Quasi-Newton method: A method that updates an approximate derivative or Jacobian.

Damping: Multiplication of a computed step by a factor less than or equal to one.

Line search: Selection of a step length to improve a merit or residual function.

Continuation: Solving a sequence of parameterized problems to reach a target solution.

Pseudo-arclength continuation: Continuation using a path parameter that handles turning points.

Conditioning: Sensitivity of a solution to perturbations in data or equations.

Scaling: Rescaling variables or residuals to improve numerical balance.

Globalization: Techniques that extend local methods to a wider convergence region.

Stopping criterion: A numerical test used to decide whether iteration should terminate.

Bifurcation: A qualitative change or branching in the set of solutions.

Trust region: A neighborhood within which a local model is considered reliable.

Review Questions

Short Answer Questions

1. What information does a sign-changing bracket provide, and what does it not prove?

2. State the local convergence condition for scalar fixed-point iteration.

3. Why is explicit inversion of a Newton Jacobian generally discouraged?

4. Distinguish a residual from an error in the unknown.

5. What is the purpose of damping a Newton step?

Long Answer Questions

1. Explain the algorithm, convergence behavior, advantages, and limitations of bisection and regula falsi.

2. Derive the Newton iteration for a scalar equation and discuss conditions that can cause it to fail.

3. Explain how quasi-Newton methods reduce the cost of solving coupled nonlinear systems.

4. Describe continuation and explain why pseudo-arclength continuation is useful near a limit point.

5. Develop a complete workflow for solving a nonlinear engineering equation, including formulation, scaling, convergence tests, and verification.

Analytical / Application Questions

1. For a nonlinear thermal balance with radiation and convection, propose a physically meaningful bracket and explain how you would combine bisection with Newton’s method.

2. A fixed-point iteration oscillates between two values. Explain two possible causes and propose computational remedies.

3. A coupled mechanical model converges according to the residual norm, but one displacement is outside its allowable range. Assess the result and specify the checks and corrective actions required.

22. Multiple Choice Questions with Answers

Question 1

Which condition guarantees at least one scalar root in an interval for a continuous function?

A. The derivative is zero throughout the interval

B. The function has the same sign at both endpoints

C. The function has opposite signs at the endpoints

D. The midpoint has the smallest function value

Correct Answer: C. The function has opposite signs at the endpoints

Question 2

What is the principal numerical advantage of bisection?

A. Quadratic convergence from any initial guess

B. Guaranteed interval reduction when its assumptions hold

C. No need for function evaluations

D. Automatic detection of every root

Correct Answer: B. Guaranteed interval reduction when its assumptions hold

Question 3

A scalar fixed-point iteration is locally convergent when which condition holds at the fixed point?

A. |g'(x*)| is greater than one

B. g'(x*) is undefined

C. |g'(x*)| is less than one

D. g'(x*) equals the residual

Correct Answer: C. |g'(x*)| is less than one

Question 4

Newton’s method may fail when the derivative at an iterate is very small because the method may

A. Produce an excessively large step

B. Guarantee bisection instead

C. Make the residual identically zero

D. Convert the problem into a linear system permanently

Correct Answer: A. Produce an excessively large step

Question 5

For a nonlinear system, the Newton correction is obtained by solving

A. J(x)Δx = F(x)

B. J(x)Δx = −F(x)

C. F(x)Δx = −J(x)

D. J(x)F(x) = Δx

Correct Answer: B. J(x)Δx = −F(x)

Question 6

What is the main purpose of variable scaling in a coupled nonlinear solve?

A. To remove all nonlinear terms

B. To make equations and variables numerically comparable

C. To guarantee uniqueness of the solution

D. To avoid evaluating residuals

Correct Answer: B. To make equations and variables numerically comparable

Question 7

A quasi-Newton method differs from classical Newton’s method primarily because it

A. Never uses function values

B. Updates an approximation to derivative information

C. Can solve only linear equations

D. Requires a sign-changing bracket

Correct Answer: B. Updates an approximation to derivative information

Question 8

Why can a small residual fail to imply a small solution error?

A. The function may be nearly flat near the root

B. The function must then be discontinuous

C. The solution is necessarily unique

D. The Jacobian is always exactly zero

Correct Answer: A. The function may be nearly flat near the root

Question 9

In continuation, the solution at one parameter value is normally used as

A. The stopping tolerance for the next problem

B. The initial guess for the next problem

C. The Jacobian determinant

D. A replacement for the residual function

Correct Answer: B. The initial guess for the next problem

Question 10

Which practice is most appropriate after a nonlinear solver reports convergence?

A. Accept the result without further checks

B. Check only the number of iterations

C. Substitute the solution into the original equations and check physical constraints

D. Increase all tolerances until the result changes

Correct Answer: C. Substitute the solution into the original equations and check physical constraints