Sign in to save

Bookmark this page so you can find it later.

Sign in to save

Bookmark this page so you can find it later.

Numerical methods turn difficult mathematical problems into step-by-step computations that can be carried out by hand, calculator, or computer. This reference covers the main tools used to approximate roots, derivatives, integrals, and solutions to differential equations. College students need it to compare methods, track error, and choose stable algorithms for applied problems.

It is especially useful when exact algebraic solutions are unavailable or inefficient.

The core ideas are approximation, convergence, and error control. Error is measured with quantities such as absolute error xx^|x - \hat{x}|, relative error xx^x\frac{|x - \hat{x}|}{|x|}, and method order. Root-finding methods such as bisection and Newton’s method approximate solutions to f(x)=0f(x)=0, while interpolation estimates function values between data points.

Numerical integration and ODE methods approximate accumulated change using formulas such as the trapezoidal rule, Simpson’s rule, Euler’s method, and Runge-Kutta methods.

Key Facts

  • Absolute error is Eabs=xx^E_{\text{abs}} = |x - \hat{x}|, where xx is the true value and x^\hat{x} is the approximation.
  • Relative error is Erel=xx^xE_{\text{rel}} = \frac{|x - \hat{x}|}{|x|} when x0x \neq 0, and percent error is 100Erel%100E_{\text{rel}}\%.
  • The bisection method requires f(a)f(b)<0f(a)f(b)<0 and uses the midpoint c=a+b2c = \frac{a+b}{2} to shrink the interval containing a root.
  • Newton’s method updates an approximation by xn+1=xnf(xn)f(xn)x_{n+1}=x_n-\frac{f(x_n)}{f'(x_n)} and can converge quickly when x0x_0 is close to a simple root.
  • The secant method avoids derivatives by using xn+1=xnf(xn)xnxn1f(xn)f(xn1)x_{n+1}=x_n-f(x_n)\frac{x_n-x_{n-1}}{f(x_n)-f(x_{n-1})}.
  • The Lagrange interpolation polynomial is Pn(x)=i=0nyijixxjxixjP_n(x)=\sum_{i=0}^{n} y_i\prod_{j\neq i}\frac{x-x_j}{x_i-x_j} for data points (xi,yi)(x_i,y_i).
  • The composite trapezoidal rule is abf(x)dxh2[f(x0)+2i=1n1f(xi)+f(xn)]\int_a^b f(x)\,dx \approx \frac{h}{2}\left[f(x_0)+2\sum_{i=1}^{n-1}f(x_i)+f(x_n)\right], where h=banh=\frac{b-a}{n}.
  • Euler’s method for y=f(t,y)y'=f(t,y) uses yn+1=yn+hf(tn,yn)y_{n+1}=y_n+h f(t_n,y_n) to approximate the solution one step at a time.

Vocabulary

Absolute Error
Absolute error is the size of the difference between the true value and an approximation, written as xx^|x-\hat{x}|.
Convergence
Convergence means that a sequence of numerical approximations approaches the exact solution or a limiting value.
Root-Finding
Root-finding is the process of approximating values of xx that satisfy f(x)=0f(x)=0.
Interpolation
Interpolation estimates a function value between known data points using a fitted function such as a polynomial.
Step Size
Step size is the spacing hh between consecutive points in a numerical method, often controlling accuracy and cost.
Stability
Stability describes whether small errors from rounding or approximation remain controlled as a numerical method proceeds.

Common Mistakes to Avoid

  • Using Newton’s method when f(xn)f'(x_n) is near 00 is a mistake because the update xn+1=xnf(xn)f(xn)x_{n+1}=x_n-\frac{f(x_n)}{f'(x_n)} can make a very large or unstable jump.
  • Applying bisection without checking f(a)f(b)<0f(a)f(b)<0 is a mistake because the method needs a sign change to guarantee a root in the interval.
  • Confusing absolute error with relative error is a mistake because xx^|x-\hat{x}| measures raw difference, while xx^x\frac{|x-\hat{x}|}{|x|} measures the error compared with the size of the true value.
  • Using a very large step size hh in Euler’s method is a mistake because the approximation yn+1=yn+hf(tn,yn)y_{n+1}=y_n+h f(t_n,y_n) can accumulate large truncation error.
  • Using a high-degree interpolation polynomial without checking behavior between points is a mistake because oscillations can occur even when the polynomial matches all data values exactly.

Practice Questions

  1. 1 If the true value is x=2.500x=2.500 and an approximation is x^=2.475\hat{x}=2.475, find the absolute error and relative error.
  2. 2 Use one step of Newton’s method for f(x)=x22f(x)=x^2-2 with initial guess x0=1.5x_0=1.5 to compute x1x_1.
  3. 3 Approximate 02x2dx\int_0^2 x^2\,dx using the trapezoidal rule with n=2n=2 subintervals.
  4. 4 Explain why a method with a smaller step size hh may be more accurate but also more computationally expensive.

Understanding Numerical Methods Reference

Every numerical result has at least two possible sources of error. Truncation error comes from replacing an infinite process with a finite one. Rounding error comes from storing only a limited number of digits.

These errors can interact in surprising ways. Taking a smaller step often reduces truncation error, but very tiny steps may make subtraction lose important digits. This is called cancellation.

A calculator can display many decimal places without making all of them trustworthy. Students should keep extra digits during intermediate work, then round only at the end. They should check whether successive answers are settling toward one value rather than trusting a single output.

Root finding is really a search for where a model changes sign or reaches a target level. A bracketing method is dependable because it keeps the root trapped inside an interval, provided the function is continuous there. Its weakness is steady but slow progress.

Methods based on tangent lines or secant lines can reach an answer in far fewer steps. Their weakness is that a poor starting estimate may send the calculation away from the desired root. A nearly horizontal graph causes trouble because the slope gives little useful direction.

Graphing the function first is often the best practical step. It reveals multiple roots, sharp turns, discontinuities, and reasonable starting values.

Interpolation uses known measurements to estimate missing values, such as temperature between weather station readings or a vehicle position between recorded times. The fitted curve passes through the data points, but that does not prove it describes the physical system well. High degree polynomials can swing wildly near the ends of an interval, even when the input data are accurate.

This effect is especially likely when data points are spaced evenly across a wide range. Using several low degree pieces, called splines, is often safer for real data. Numerical differentiation has an extra difficulty.

It estimates slope by subtracting nearby function values, so small measurement errors can become much larger. Smooth data and sensible spacing matter greatly.

Integration methods estimate total quantity from sampled values. They appear in work calculations, distance from velocity data, total electric charge from current data, and areas under experimental graphs. A rule that uses straight segments works well when the graph bends gently.

A rule that uses curved pieces can be more accurate for smooth functions, yet it still depends on the interval width and the behavior between sample points. Differential equation methods extend this idea through time. Each new estimate becomes the starting point for the next one, so early errors can spread forward.

This matters in population models, falling objects, circuits, and disease models. Students should compare results using different step sizes.

If halving the step changes the answer greatly, the original calculation was not reliable enough. Some systems are stable only with small steps, so choosing a method is about reliability as much as speed.