Numerical Methods, Vectors, and Engineering Models
Key Takeaways
- FE numerical-methods items ask for one or two iterations, not a full algorithm: Newton's method x_{n+1} = x_n − f(x_n)/f'(x_n).
- Numerical integration uses the trapezoidal rule and Simpson's 1/3 rule (Simpson needs an even number of equal intervals).
- Vectors must be resolved into i, j, k components before applying equilibrium, dot products, cross products, or work.
- The dot product A·B = |A||B|cosθ gives projections and work; the cross product A×B = |A||B|sinθ gives moments and areas.
- Engineering model selection means naming the governing equation before substituting numbers, then performing a reasonableness check.
Numerical Methods at FE Level
FE numerical-methods questions test whether you can run one or two steps of a standard algorithm by hand, not implement software. The two patterns that appear most are root-finding and numerical integration.
Newton–Raphson Root Finding
Newton's method iterates x_{n+1} = x_n − f(x_n)/f'(x_n), converging quadratically near a simple root. Given f(x) = x² − 10 and x₀ = 3: f(3) = −1, f'(3) = 6, so x₁ = 3 − (−1)/6 = 3.167—already close to √10 ≈ 3.162. The exam typically asks for x₁ after one iteration, so know the formula and substitute carefully.
The bisection method halves a bracketing interval [a, b] where f(a) and f(b) have opposite signs; the new midpoint replaces whichever endpoint keeps the sign change. It is slower but guaranteed to converge.
Numerical Integration
When a function is known only at data points, approximate ∫f dx numerically.
Trapezoidal rule: ∫ₐᵇ f dx ≈ (h/2)[f₀ + 2f₁ + 2f₂ + … + 2f_{n−1} + f_n], where h = (b − a)/n is the equal step size. Each interval is treated as a trapezoid.
Simpson's 1/3 rule: ∫ₐᵇ f dx ≈ (h/3)[f₀ + 4f₁ + 2f₂ + 4f₃ + … + 4f_{n−1} + f_n]. It fits parabolas and is more accurate, but requires an even number of equal intervals. A classic trap is applying Simpson's rule with an odd interval count—check the count first.
Worked Example — Trapezoidal
Velocity data at t = 0, 1, 2 s is v = 0, 4, 6 m/s. Distance ≈ (1/2)[0 + 2(4) + 6] = (1/2)(14) = 7 m. With Simpson's 1/3 (n = 2, even): (1/3)[0 + 4(4) + 6] = (1/3)(22) = 7.33 m.
Vector Operations
Mechanics is built on vectors. Always resolve into components F_x = F cosθ, F_y = F sinθ before summing. The resultant magnitude is √(F_x² + F_y²) at angle θ = tan⁻¹(F_y/F_x).
The dot product A·B = A_xB_x + A_yB_y + A_zB_z = |A||B|cosθ yields projections and work W = F·d. The cross product A×B has magnitude |A||B|sinθ, is perpendicular to both vectors by the right-hand rule, and gives moments M = r×F and parallelogram areas.
Engineering Model Selection and Computational Tools
The highest-value habit on the FE is naming the governing relationship before plugging numbers: "this is a moment balance," "this is conservation of energy," "this is a first-order decay." Writing the equation first prevents mixing incompatible formulas and makes unit checking trivial.
Computational-tools questions test spreadsheet and pseudo-code literacy: cell references (relative A1 vs. absolute $A$1), range functions like =SUM(A1:A10) and =AVERAGE(B1:B5), and simple loop logic where a variable updates each pass. You may be asked to predict a cell's value after a formula is filled down, or to identify an off-by-one error in a counter.
| Method | When to use | FE tip |
|---|---|---|
| Newton–Raphson | Smooth f with known f' | One iteration usually asked |
| Bisection | Sign change bracketed | Always converges |
| Trapezoidal | Tabulated data | Any interval count |
| Simpson's 1/3 | Tabulated data | Even intervals only |
| Dot product | Work, projection | Scalar result |
| Cross product | Moment, area | Vector result |
Close every computation with a reasonableness check: correct order of magnitude, sensible sign, and consistent units. A negative distance or a moment ten times too large signals a setup error you can fix before submitting.
Interpolation, Finite Differences, and Error Awareness
Two more numerical tools round out the FE set. Linear interpolation estimates a value between two tabulated points—essential when reading steam tables or material-property charts. The formula is y = y₀ + (x − x₀)·(y₁ − y₀)/(x₁ − x₀). For instance, if a property is 40 at T = 100 and 60 at T = 200, the value at T = 150 is 40 + (50/100)(20) = 50. The exam frequently buries this inside a thermodynamics or heat-transfer question rather than labeling it "interpolation."
Finite differences approximate derivatives from data: the forward difference f'(x) ≈ [f(x+h) − f(x)]/h, and the more accurate central difference f'(x) ≈ [f(x+h) − f(x−h)]/(2h). These let you estimate velocity from position data or a slope from a table without a closed-form function.
Sources of Numerical Error
Every numerical method carries error, and the FE may ask you to reason about it qualitatively:
- Round-off error accumulates when intermediate values are truncated—mitigate by keeping full precision until the final step.
- Truncation error comes from approximating an infinite process with a finite one (fewer iterations, larger step size h). Halving h generally reduces trapezoidal error by about a factor of four.
- Convergence matters: Newton's method converges fast but can diverge if f'(x) is near zero or the initial guess is poor; bisection is slow but always converges within a bracket.
The practical takeaway is to pick the method that fits the data you are given, run only the iterations asked for, keep full precision, and finish with a magnitude check. Overbuilding—running five iterations when one is requested, or coding a full algorithm—wastes the scarce time the FE allots (about 3 minutes per question).
Using Newton's method on f(x) = x² − 5 with x₀ = 2, what is x₁?
Simpson's 1/3 rule can be applied directly only when the number of equal intervals is:
A 50 N force acts at 30° above horizontal. Its horizontal component is closest to:
A tabulated property is 30 at x = 10 and 50 at x = 20. By linear interpolation, the value at x = 14 is: