Not Every Good Method Needs an Exact Decimal Answer
Suppose we want to solve
\[ x^2=2. \]An exact symbolic answer is
\[ x=\pm\sqrt2. \]But if we need a decimal value for measurement or computation, we might use an approximation such as
\[ \sqrt2\approx1.4142. \]Now notice the important distinction:
- \(\sqrt2\) is an exact value.
- \(1.4142\) is an approximate value.
- A procedure that repeatedly improves \(1.4,1.41,1.414,\ldots\) is an approximate method.
What Is an Exact Method?
An exact method is a procedure that, under its stated assumptions, produces the mathematically exact result rather than an approximation.
Example: Greatest Common Divisor
Euclid's algorithm computes the exact greatest common divisor of two positive integers.
For \(48\) and \(18\):
\[ 48=2(18)+12, \] \[ 18=1(12)+6, \] \[ 12=2(6)+0. \]Therefore,
\[ \boxed{\gcd(48,18)=6}. \]The answer \(6\) is exact.
What Is an Approximate Method?
An approximate method is designed to produce a value near the desired result, usually with a measurable or controllable error.
Approximate methods are useful when:
- an exact form is unavailable,
- an exact form is too expensive to compute,
- measured data is already approximate,
- only a specified precision is needed.
The goal is not “close enough by guesswork.”
A good approximation method gives a systematic way to improve or bound the error.
Method versus Result
An exact method can still be followed by numerical approximation.
For example, the quadratic formula gives
\[ x=\frac{-b\pm\sqrt{b^2-4ac}}{2a} \]exactly.
If the radical is irrational, the final decimal may still be approximate.
Conversely, an approximate method may happen to land exactly on the answer for a particular input.
So keep these ideas separate:
Exact Method
The procedure is guaranteed to produce the exact mathematical result.
Approximate Method
The procedure seeks a sufficiently close result with controlled error.
Exact Value
A representation such as \(\sqrt2\), \(1/3\), or \(6\).
Approximate Value
A rounded or estimated representation such as \(1.4142\).
Finite Procedures
Many exact algorithms terminate after a finite number of steps.
For example, solving
\[ 3x+9=24 \]symbolically requires only:
\[ 3x=15, \] \[ x=5. \]No refinement loop is needed.
Approximate methods may also terminate after finitely many steps—but they stop because the approximation has become sufficiently good, not because an exact representation has necessarily been reached.
Successive Refinement
A common approximate strategy begins with a rough estimate and repeatedly improves it.
For \(\sqrt2\), one possible sequence is
\[ 1.4,\quad1.41,\quad1.414,\quad1.4142,\ldots \]Each new value is intended to reduce the error.
This general pattern is called successive refinement.
\[ \boxed{ x_0 \to x_1 \to x_2 \to \cdots \to \text{better approximation} } \]Convergence
A sequence of approximations
\[ x_0,x_1,x_2,\ldots \]converges to a value \(L\) when the approximations can be made arbitrarily close to \(L\) by continuing sufficiently far.
Informally:
\[ x_n\to L. \]Not every iterative procedure converges. Some oscillate, diverge, or stall.
Stopping Criteria
Approximation methods need a rule for deciding when enough refinement has been performed.
Common stopping criteria include:
- a known error bound is below a required tolerance,
- successive approximations differ by less than a threshold,
- a residual such as \(|f(x)|\) is sufficiently small,
- a maximum number of iterations has been reached.
Example
Stop when
\[ |x_{n+1}-x_n|<10^{-5}. \]This controls the change between successive estimates, though the difference between successive estimates is not always identical to the true error.
Error Tolerance
An error tolerance is a maximum acceptable error for the purpose of the computation.
For example, we may require
\[ |x_{\text{approx}}-x_{\text{true}}|<0.001. \]Then any approximation meeting that condition is accurate enough for the specified task.
A tighter tolerance generally requires more work.
\[ \boxed{ \text{smaller allowed error} \Longrightarrow \text{usually more computation} } \]Upper and Lower Bounds
Approximation becomes especially trustworthy when the true value is trapped between known bounds.
Suppose
\[ 1.414<\sqrt2<1.415. \]Then the interval width is
\[ 1.415-1.414=0.001. \]The midpoint is
\[ 1.4145. \]The midpoint error is at most half the interval width:
\[ \boxed{0.0005}. \]Bounds give a direct mathematical guarantee rather than relying only on the appearance of convergence.
Bisection as an Approximate Method
Bisection finds a root by repeatedly cutting a sign-changing interval in half.
Suppose
\[ f(a)f(b)<0 \]and \(f\) is continuous on \([a,b]\).
Then a root lies between \(a\) and \(b\).
The algorithm:
- Compute the midpoint \(m=(a+b)/2\).
- Determine which half still contains a sign change.
- Keep that half.
- Repeat.
Every iteration halves the interval width.
After \(n\) halvings, the width is
\[ \boxed{ \frac{b-a}{2^n} }. \]That gives a simple, predictable error control.
Newton-Style Refinement
Some methods update an estimate using local information about the function.
Newton's method uses
\[ \boxed{ x_{n+1} = x_n-\frac{f(x_n)}{f'(x_n)} } \]when the derivative is available and the method is used under suitable conditions.
For
\[ f(x)=x^2-2, \]we have
\[ f'(x)=2x, \]so
\[ x_{n+1} = x_n-\frac{x_n^2-2}{2x_n} = \frac12\left(x_n+\frac2{x_n}\right). \]Starting from \(x_0=1.5\), the estimates quickly approach \(\sqrt2\).
Unlike bisection, Newton's method is not guaranteed to behave well from every starting value. Faster methods may require stronger assumptions.
Exact Arithmetic and Finite Precision
A mathematically exact method may still be implemented on a machine that uses finite-precision arithmetic.
For example,
\[ \frac13 \]is exact symbolically, but a floating-point system may store only a finite approximation such as
\[ 0.3333333333. \]So distinguish:
- the exactness of the mathematical method,
- the exactness of the representation used by the machine.
Computer algebra systems can sometimes preserve exact fractions and radicals, while ordinary numerical tools often approximate them.
Tradeoffs among Methods
| Method Feature | Possible Advantage | Possible Cost |
|---|---|---|
| Exact symbolic method | Preserves structure and exactness | May be algebraically difficult or unavailable |
| Guaranteed bounded approximation | Clear error control | May converge slowly |
| Fast iterative approximation | Can reach high accuracy quickly | May depend on starting values or assumptions |
| Low-precision approximation | Cheap and fast | May be insufficient for the task |
There is no universally best method.
The right method balances:
- accuracy,
- speed,
- reliability,
- available information,
- the mathematical structure of the problem.
Choosing an Appropriate Method
Ask:
- Is an exact answer available and useful?
- How accurate does the answer need to be?
- Can I bound the error?
- Does the method always terminate?
- Does the approximation converge?
- How much computation is practical?
- Are the inputs already measured or approximate?
A decimal approximation to twelve places may be wasted effort if the measured inputs are only known to three significant figures.
Likewise, insisting on a complicated exact form may be unnecessary when a four-decimal answer is all the application needs.
Practical Applications
Engineering
Approximate methods solve equations and models that have no convenient exact solution.
Science
Measured inputs are combined with numerical models and tolerance requirements.
Computer Graphics
Curves, lighting, and geometry are often approximated to visually sufficient precision.
Finance
Iterative methods may solve for interest rates or yields that cannot be isolated conveniently by elementary algebra.
Mathematics
Exact proofs and symbolic derivations coexist with numerical approximation, root-finding, integration, and optimization.
Worked Examples
Example 1: Exact Linear Solution
\[ 5x-7=18. \]Add \(7\):
\[ 5x=25. \]Divide by \(5\):
\[ \boxed{x=5}. \]This is an exact finite method.
Example 2: Exact Form, Approximate Decimal
\[ x^2=7. \]Exact:
\[ \boxed{x=\pm\sqrt7}. \]Approximate:
\[ x\approx\boxed{\pm2.6458}. \]Example 3: Bounding \(\sqrt2\)
Since
\[ 1.41^2=1.9881<2 \]and
\[ 1.42^2=2.0164>2, \]we know
\[ \boxed{ 1.41<\sqrt2<1.42 }. \]Example 4: Bisection Width
Suppose the initial root-containing interval has width
\[ 1. \]After \(8\) bisections, the width is
\[ \frac1{2^8} = \frac1{256} \approx0.00390625. \]So the root is trapped in an interval narrower than \(0.004\).
Example 5: Successive Refinement
Use
\[ x_{n+1} = \frac12\left(x_n+\frac2{x_n}\right) \]with \(x_0=1.5\).
\[ x_1 = \frac12\left(1.5+\frac2{1.5}\right) \approx1.41667, \] \[ x_2 \approx1.41422. \]The approximation rapidly approaches
\[ \sqrt2\approx1.41421. \]Example 6: Stop by Tolerance
Suppose an algorithm produces
\[ 3.1414,\quad3.14158,\quad3.141592. \]If the required tolerance is roughly \(10^{-4}\), continuing many more iterations may be unnecessary for the stated task.
Common Errors
1. Calling Every Decimal Method Approximate
Some decimal values, such as \(0.5\), represent exact numbers. The method and representation must be considered separately.
2. Assuming an Approximate Method Is Unreliable
Approximation methods can come with rigorous convergence proofs and error bounds.
3. Stopping Because Consecutive Values Look Similar
Small changes between iterates do not always prove small error relative to the true solution.
4. Ignoring the Stopping Rule
An iterative algorithm needs a defined criterion for deciding when enough work has been done.
5. Demanding More Precision Than the Inputs Support
Measured data can limit meaningful output precision.
6. Assuming Faster Convergence Always Means a Better Method
A faster method may be less robust or require stronger assumptions.
7. Confusing Exact Mathematical Arithmetic with Machine Arithmetic
A mathematically exact fraction may be stored approximately in floating-point form.
Practice
Exact or Approximate Method?
Solving
\[ 2x+3=11 \]by inverse operations is usually what kind of method?
Check Answer
\[ \boxed{\text{exact method}} \]Approximation Bound
If
\[ 4.12Check Answer
\[ 4.14-4.12 = \boxed{0.02}. \]Midpoint Error Bound
Using the previous interval, what is the largest possible error of the midpoint \(4.13\)?
Check Answer
\[ \boxed{0.01}. \]Bisection
If an interval has width \(8\), what is its width after three bisections?
Check Answer
\[ \frac8{2^3} = \boxed1. \]Method Choice
If a physical measurement is only reliable to the nearest \(0.1\) unit, is there usually value in computing a model output to fifteen decimal places?
Check Answer
Usually no. The meaningful precision of the result is limited by the quality of the inputs and the needs of the application.
The Deeper Idea
Exact and approximate methods answer different computational needs.
\[ \boxed{ \text{exact method} \to \text{mathematically exact result} } \] \[ \boxed{ \text{approximate method} \to \text{controlled closeness to a desired result} } \]Approximation becomes rigorous when we can describe:
- what quantity is being approximated,
- how the estimate is improved,
- why the process converges,
- how large the remaining error can be,
- when the process should stop.
This prepares the way for later numerical analysis, where approximation is not treated as a compromise but as a mathematical discipline with its own proofs, bounds, and algorithms.
Summary
- An exact method produces the mathematically exact result under its stated assumptions.
- An approximate method produces a sufficiently close result with controlled or estimated error.
- The exactness of a method is different from the exactness of a displayed value.
- Successive refinement generates increasingly improved approximations.
- A convergent method approaches a limiting value as the process continues.
- Stopping criteria determine when an approximation is sufficiently good.
- Error tolerance states the maximum acceptable error for a task.
- Upper and lower bounds can trap the true value and provide guaranteed error control.
- Bisection repeatedly halves a root-containing interval.
- Newton-style methods can converge rapidly but require stronger conditions.
- A mathematically exact algorithm may still be implemented with approximate machine arithmetic.
- Method choice balances accuracy, speed, robustness, available information, and the purpose of the computation.
Exact methods aim to finish with the exact mathematical result. Approximate methods aim to control the distance from that result. Both are rigorous when their assumptions, guarantees, and stopping rules are understood.
Algorithms and Computational Thinking
- Previous: Symbolic and Numerical Computation
- Algorithms and Computational Thinking — return to the lesson directory.