Mathematical approximation is the disciplined process of replacing an exact quantity, relationship or problem with a nearby representation that is easier to calculate, understand, measure or use—while keeping control of the error introduced. It is one of the most practical ideas in mathematics because exact answers are often unavailable, unnecessary, expensive to compute, impossible to measure, or less useful than a well-understood approximation.
Approximation is not careless guessing. Good approximation has a declared target, a reason for simplifying, a known or estimated error, a domain in which the approximation is trustworthy, and a decision about whether the remaining error is acceptable for the job.
This page belongs to the wider How Mathematics Works library. It complements How Mathematics Works | Numerical Analysis, while keeping a broader reader job: how estimation, rounding, error control, simplification and approximation operate from school mathematics through modelling, engineering, science and computation.
The Short Answer
Mathematical approximation works through a controlled trade:
exact or difficult object → choose a simpler representation → calculate or estimate → measure or bound the error → check whether the error is acceptable → use the result only inside its valid range
The strongest approximation is not necessarily the closest possible one. It is the approximation that is sufficiently accurate for the purpose, computationally reasonable, transparent about its limitations and stable enough to support the next decision.
1. Exact Mathematics and Approximate Mathematics Are Partners
Mathematics is often introduced as the search for exact answers. Sometimes exactness is essential. If an algebraic identity is claimed, approximate equality is not enough. If a theorem is being proved, logical validity is not negotiable.
But many useful mathematical tasks concern measured quantities, complicated equations, irrational numbers, enormous data sets, differential equations, simulations or optimisation problems where exact closed-form answers are unavailable or operationally unnecessary.
Exact mathematics provides the structure. Approximation provides a way to work when the exact structure cannot be evaluated directly or does not need to be.
2. Why Approximation Exists
- Measurement is finite. A ruler, sensor or survey does not return infinitely precise truth.
- Some numbers are irrational. π and √2 cannot be written as terminating or repeating decimal expansions.
- Some equations have no convenient closed form. Numerical methods are needed.
- Some exact calculations are too expensive. A fast approximation may be operationally superior.
- Models simplify reality. Their outputs inherit modelling uncertainty.
- Decisions have tolerances. A rough estimate may be sufficient for an early screening decision.
- Communication requires compression. Reporting 3.141592653589793 may be less useful than reporting 3.14.
The point is not to abandon accuracy. It is to match accuracy to purpose.
3. Approximation Begins With a Tolerance
Before approximating, ask how much error the task can tolerate.
A grocery estimate might tolerate a dollar of error. A building component may need millimetre control. A navigation system may need metre-level accuracy. A spacecraft manoeuvre may demand much tighter modelling and measurement tolerances.
The required accuracy comes from the consequences of error, not from the number of decimal places a calculator can display.
4. Estimation: The Simplest Approximation Tool
Estimation replaces exact inputs with simpler nearby values so a result can be obtained quickly.
For example:
19.8 × 5.1 ≈ 20 × 5 = 100.
The estimate is not intended to replace the exact product in every context. It gives a scale check. If a calculator later returns 1008, the estimate immediately reveals that a decimal-place error is likely.
5. Estimation as Error Detection
Estimation is often taught as a separate topic, but one of its deepest uses is verification. Before accepting a precise-looking answer, ask whether its order of magnitude is plausible.
If 49.7 × 0.21 is calculated as 1043.7, a rough estimate of 50 × 0.2 = 10 immediately exposes the answer as unreasonable.
Approximation therefore protects exact calculation from silent mistakes.
6. Rounding
Rounding replaces a number with a nearby value at a chosen place value or precision.
12.746 rounded to 2 decimal places = 12.75 12.746 rounded to 3 significant figures = 12.7
The two answers are different because decimal places and significant figures answer different precision questions.
7. Decimal Places and Significant Figures
Decimal places count positions after the decimal point. Significant figures count meaningful digits beginning from the first non-zero digit.
For 0.004862:
3 significant figures = 0.00486 3 decimal places = 0.005
The choice should follow the context. Engineering, science and measurement often care about significant figures because they preserve relative precision across scales.
8. Rounding Creates an Interval, Not Just a Number
If a length is reported as 8.4 cm correct to the nearest 0.1 cm, the true measured value is understood to lie in an interval:
8.35 ≤ length < 8.45.
The rounded value 8.4 is therefore shorthand for a range of possible underlying values. Error bounds make that hidden range explicit.
9. Absolute Error
Absolute error measures the size of the difference between an approximate value and the true or reference value:
absolute error = |approximation - true value|.
If √2 is approximated by 1.414, then the absolute error is about 0.000214.
Absolute error is easy to interpret in the original units, but it does not tell us how large the error is relative to the quantity itself.
10. Relative Error
Relative error compares the absolute error with the size of the true or reference value:
relative error = absolute error / |true value|.
An error of 1 metre is enormous when measuring a 2 metre object and negligible when measuring a 1,000 kilometre route. Relative error captures that scale dependence.
11. Percentage Error
Percentage error expresses relative error as a percentage:
percentage error = relative error × 100%.
This is especially useful when comparing accuracy across quantities with different magnitudes.
12. Upper and Lower Bounds
A bound does not claim the exact value. It guarantees that the value lies above, below or between stated limits.
Bounds are often more valuable than a single estimate because they preserve what is known about uncertainty.
lower bound ≤ true value < upper bound
A result such as “the cost will be between $9.8 million and $10.4 million under these assumptions” may support planning better than a falsely precise point estimate of $10,031,472.18.
13. Interval Arithmetic
If inputs are known only within intervals, calculations can propagate those ranges.
Suppose length L lies between 9.95 and 10.05 metres, and width W lies between 4.95 and 5.05 metres. A conservative rectangular area interval can be formed from the appropriate endpoint products.
This keeps uncertainty visible instead of pretending that rounded input values were exact.
14. Error Propagation
Approximate inputs create approximate outputs. The important question is how input error travels through a calculation.
If two measured lengths are added, their absolute uncertainties can accumulate. In multiplication, relative uncertainties often become more informative. For nonlinear functions, the output sensitivity can vary dramatically with the input region.
Error propagation is the mathematics of not forgetting where uncertainty came from.
15. Sensitivity
A problem is sensitive when small input changes can create large output changes.
Consider the function:
f(x) = 1 / (x - 1).
Near x = 1, tiny input changes produce huge output changes. Even excellent measurement precision may not give a stable output there.
This is different from a bad numerical algorithm. The underlying mathematical problem itself may be ill-conditioned or sensitive.
16. Condition and Numerical Error Are Different
Two distinct questions should be separated:
- Problem conditioning: how much does the true answer change when the input changes slightly?
- Algorithm stability: how much additional error does the computational method introduce?
A stable algorithm cannot make an inherently sensitive problem insensitive. Conversely, a well-conditioned problem can still be damaged by a poor numerical method.
17. Round-Off Error
Computers store numbers with finite precision. Many real numbers cannot be represented exactly in a finite binary format. Every rounding operation can therefore introduce a small discrepancy.
Usually these errors are tiny. In long or poorly conditioned calculations they can accumulate or amplify.
Numerical mathematics studies how to keep these errors under control.
18. Truncation Error
Truncation error appears when an infinite or continuous mathematical process is replaced by a finite approximation.
- replacing an infinite series with its first few terms;
- approximating a derivative using a finite difference;
- approximating an integral with finitely many strips;
- advancing a differential equation through finite time steps.
Round-off error comes from finite number representation. Truncation error comes from simplifying the mathematical procedure itself.
19. Approximate Equality
The symbol ≈ should be used when two quantities are close but not exactly equal.
π ≈ 3.14159 √2 ≈ 1.41421
Using = instead of ≈ silently converts an approximation into a false exact statement. Mathematical notation should preserve the status of the claim.
20. Rational Approximations to Irrational Numbers
Irrational numbers cannot be represented exactly as ratios of integers, but useful rational approximations can be extremely accurate.
For π:
22/7 ≈ 3.142857... 355/113 ≈ 3.14159292...
The second approximation is remarkably close for such modest integers. Continued fractions provide a systematic framework for finding strong rational approximations to real numbers.
21. Approximation by Local Linearity
Smooth curves look almost straight when examined closely enough. Calculus turns this observation into a method.
Near x = a:
f(x) ≈ f(a) + f'(a)(x - a).
This is the linear approximation or tangent-line approximation. It replaces a potentially complicated function with a nearby line.
22. Worked Example: Estimating √4.04
Let f(x) = √x and choose a = 4. Then:
f(4) = 2 f'(x) = 1/(2√x) f'(4) = 1/4.
For x = 4.04:
√4.04 ≈ 2 + (1/4)(0.04)
= 2.01.
The exact calculator value is close to 2.00998. The linear approximation captures it extremely well because 4.04 is near the expansion point 4.
23. Why Linearisation Has a Local Domain
A tangent line approximates a curve well near the contact point. Farther away, curvature matters more and the linear model can drift badly.
This illustrates a general law of approximation: a method can be excellent inside one region and poor outside it. Every approximation should carry an operating range.
24. Taylor Approximation
Taylor polynomials extend linearisation by matching more derivatives at a chosen point.
f(x) ≈ f(a)
+ f'(a)(x-a)
+ f''(a)(x-a)²/2!
+ ...
Adding more terms can improve local accuracy when the function and expansion conditions support it. The remaining error can often be analysed mathematically.
25. Small-Angle Approximations
For angles measured in radians and sufficiently close to zero:
sin θ ≈ θ cos θ ≈ 1 - θ²/2 tan θ ≈ θ.
These approximations simplify many physical and engineering models. Their usefulness depends on the angle being small enough for the required accuracy.
The phrase “small angle” is not a universal number. The acceptable range depends on the permitted error.
26. Series Approximation as a Ladder of Accuracy
Series methods allow a modeller to choose how much accuracy to purchase with additional computation.
one term → rough approximation more terms → usually better local approximation many terms → higher cost and possible numerical issues
This creates an adjustable accuracy budget rather than an all-or-nothing choice between exact and approximate.
27. Root Finding
Many equations cannot be rearranged into a convenient exact formula. Numerical root-finding methods search for values where:
f(x) = 0.
The answer is usually a sequence of increasingly accurate approximations rather than one symbolic expression.
28. Bisection Method
The bisection method starts with an interval whose endpoints give opposite signs for a continuous function. It repeatedly halves the interval and keeps the half that still contains a sign change.
known bracket → halve → choose surviving bracket → halve again → continue until interval is sufficiently small
Bisection is not always fast, but it is attractive because the bracket provides a transparent error bound.
29. Newton’s Method
Newton’s method uses tangent lines to generate improved root estimates:
xₙ₊₁ = xₙ - f(xₙ)/f'(xₙ).
For √2, solve f(x)=x²-2=0. Starting from x₀=1.5 gives approximately:
1.500000000 1.416666667 1.414215686 1.414213562
The convergence is rapid in this favourable case. But Newton’s method can fail or behave badly with poor starting points, zero derivatives or unsuitable functions.
30. Fast Does Not Mean Safe Everywhere
Approximation algorithms have different reliability profiles. Bisection is slower but maintains a bracket. Newton’s method can be very fast but may jump away from the desired root.
Method selection therefore balances speed, guarantees, input information and problem structure.
31. Numerical Differentiation
If an analytic derivative is unavailable but function values are known, a derivative can be approximated using finite differences.
f'(x) ≈ [f(x+h) - f(x)] / h.
Making h smaller can reduce truncation error, but if h becomes too small in finite-precision computation, subtractive cancellation and round-off can become significant.
This is a classic approximation trade-off: smaller step size is not automatically better forever.
32. Numerical Integration
Definite integrals can be approximated by replacing a curved area with simpler geometric pieces.
- rectangle rules;
- trapezoidal rule;
- Simpson-type polynomial approximations;
- Gaussian quadrature and more advanced methods.
The core idea is the same: replace a difficult continuous object with finitely many easier pieces while controlling the integration error.
33. Worked Example: Trapezoidal Approximation
Consider:
∫₀¹ x² dx.
Using two equal intervals of width 0.5, the trapezoidal rule gives:
0.5[(0 + 1)/2 + 0.25] = 0.375.
The exact integral is 1/3 ≈ 0.3333. The approximation is useful but visibly imperfect. Refining the partition improves the result for this smooth function.
34. Interpolation
Interpolation constructs a function that passes through known data points and estimates values between them.
Linear interpolation connects nearby points with straight lines. Polynomial interpolation uses higher-degree curves. Splines join lower-degree polynomials smoothly across intervals.
Interpolation is usually safer inside the observed range than extrapolation beyond it.
35. Interpolation and Regression Are Not the Same
Interpolation typically aims to pass through the supplied data points. Regression aims to fit a relationship to noisy data and does not usually pass through every point.
If measurements contain noise, forcing an interpolating curve through every observation may reproduce noise rather than underlying structure.
The approximation method should match what the data represent.
36. Piecewise Approximation
A complicated function can often be approximated by simpler rules on separate regions.
Piecewise linear models are common because each local segment is easy to compute and interpret. Splines improve smoothness while preserving local control.
Breaking one global problem into local approximations can be more stable than forcing one enormous formula to fit everything.
37. Approximation by Discretisation
Continuous systems are often approximated on discrete grids or time steps.
- a fluid field becomes values at mesh points;
- a continuous time process becomes steps Δt apart;
- a curved structure becomes finite elements;
- a spatial domain becomes cells or nodes.
The discretisation resolution determines a trade-off between computational cost and approximation error.
38. Differential Equations and Step Size
Many differential equations are solved numerically by advancing the state through small time increments.
Euler’s method uses the current slope to project forward:
yₙ₊₁ = yₙ + h f(tₙ, yₙ).
Smaller h usually improves the local approximation but increases the number of steps. Higher-order methods such as Runge–Kutta schemes use more slope information to gain accuracy per step.
39. Convergence
An approximation method converges when its approximate result approaches the desired mathematical object as the controlling resolution improves appropriately.
For example, a sequence of approximations Aₙ may satisfy:
Aₙ → A as n → ∞.
Convergence provides a theoretical reason to trust refinement. Without convergence, doing “more” computation may not move toward the correct answer.
40. Rate of Convergence
Two methods may both converge but at very different speeds. One may halve the error each iteration. Another may square the error near the solution.
Rate of convergence matters because accuracy has a computational price. A rapidly convergent method can reach a target tolerance with far fewer iterations.
41. Stopping Criteria
An iterative approximation needs a rule for when to stop.
- the change between iterations is below a threshold;
- the residual is sufficiently small;
- a certified error bound is below tolerance;
- the iteration limit is reached;
- additional improvement is no longer worth the computational cost.
Stopping too early wastes accuracy. Stopping too late wastes resources and may create the illusion that more digits are meaningful than the input uncertainty supports.
42. Residual and Error Are Different
A residual measures how well an approximate answer satisfies the original equation. Error measures how far the approximation is from the true solution.
A small residual often suggests a good solution, but for badly conditioned problems a small residual may coexist with a significant solution error.
This distinction prevents a common mistake: assuming that an equation that is almost satisfied must imply an answer that is almost correct.
43. Asymptotic Approximation
Asymptotic approximation describes how a quantity behaves when some parameter becomes very large, very small or approaches a limiting value.
Instead of preserving every detail, the approximation keeps the dominant behaviour.
This is powerful in probability, combinatorics, physics and algorithm analysis because complex formulas can simplify dramatically in limiting regimes.
44. Big-O Notation and Error Scale
Big-O notation can describe the scale of a neglected term or computational growth.
f(h) = A + O(h²)
informally indicates that the remaining error is controlled at the scale of h² as h becomes small, under the relevant conditions.
The notation does not provide a universal numerical error by itself. Constants, domains and assumptions still matter.
45. Perturbation
Perturbation methods begin with a simpler problem whose solution is known and then study how the solution changes when a small parameter modifies the system.
easy reference problem + small change ε → correction terms → improved approximation
This idea appears across mechanics, quantum theory, fluid dynamics and dynamical systems. It formalises the intuition that a slightly changed problem may have a slightly changed solution—when the system is sufficiently regular.
46. Approximation in Geometry
Geometry uses approximation whenever curved objects are represented by simpler pieces.
- a circle can be approximated by a polygon with many sides;
- a curved surface can be triangulated;
- a coastline can be represented at different map resolutions;
- a smooth shape can be replaced by finite elements for structural analysis.
Increasing resolution can improve geometric fidelity, but it also increases storage and computation.
47. Approximation in Probability
Probability contains many useful approximations. A binomial distribution under appropriate conditions may be approximated by a normal or Poisson distribution. Complicated sampling distributions may be approximated asymptotically.
The value comes from replacing a difficult calculation with a simpler one while preserving enough of the probability structure for the task.
The conditions matter. Applying an approximation outside its regime can give misleading tail probabilities or confidence statements.
48. Approximation in Statistics
Statistical estimates are approximate descriptions of unknown population quantities. Sample means, regression coefficients and confidence intervals all carry sampling uncertainty.
Approximation in statistics is therefore not merely numerical. It is epistemic: the data provide limited information about a larger population or process.
A statistical result should communicate uncertainty rather than hiding it behind a point estimate.
49. Approximation in Mathematical Modelling
Every mathematical model is an approximation of reality because it includes selected mechanisms and ignores others.
The question is not whether the model is perfectly realistic. No useful model contains every detail. The question is whether the omitted details are small enough relative to the reader job or decision.
See How Mathematical Modelling Works for the full model-validation cycle.
50. Approximation in Optimisation
Some optimisation problems are too large for an exact solution within operational time. Approximation algorithms and heuristics may seek solutions that are provably near-optimal or practically strong.
A delivery system may prefer a route found in seconds that is within a small percentage of the best possible route rather than waiting hours for exact optimality.
Approximation converts computational difficulty into a controlled performance trade.
51. Approximation in Engineering
Engineering is full of approximations because physical systems are complex and measurements are finite.
- linear material behaviour over a limited range;
- small-deflection assumptions;
- lumped-parameter electrical models;
- finite-element meshes;
- idealised boundary conditions;
- empirical correlations.
Engineering discipline lies in validating whether those approximations remain conservative or sufficiently accurate for the design case.
52. Approximation in Physics
Physics frequently begins with idealised models: point masses, frictionless surfaces, ideal gases, small oscillations or isolated systems.
These are not necessarily naive. They isolate a dominant mechanism. More effects can be added when the required accuracy demands them.
Approximation allows theory to be layered rather than attempting impossible total realism from the start.
53. Approximation in Computing
Computing uses approximation to reduce time, memory, bandwidth and energy consumption.
- lossy image and audio compression;
- approximate nearest-neighbour search;
- sampling rather than exhaustive processing;
- reduced numerical precision;
- sketching and probabilistic data structures;
- approximate optimisation.
The correct question is not whether information was discarded. It is whether the discarded information mattered for the receiver’s task.
54. Approximation in Finance
Financial models approximate uncertain future cash flows, risks, correlations and market behaviour. Even sophisticated models cannot remove uncertainty from future events.
A model price may be numerically precise while resting on approximate volatility, liquidity and behavioural assumptions.
Responsible use separates computational precision from economic certainty.
55. Approximation and Measurement
Measured values are already approximations before any mathematics is performed.
A reported mass of 2.37 kg may depend on instrument resolution, calibration, environmental conditions and measurement procedure. Carrying that value through twelve decimal places in later calculations does not create twelve-decimal-place knowledge.
The output cannot inherit more meaningful precision than the measurement process supports.
56. False Precision
False precision occurs when a result is reported with more numerical detail than the underlying information justifies.
A population forecast of 5,214,782 people may look authoritative, but if the modelling uncertainty is ±100,000, the final digits have little decision value.
Precision in display is not the same as accuracy in knowledge.
57. Rounding Too Early
Repeated rounding during intermediate steps can accumulate avoidable error.
A common practical rule is to retain sufficient guard digits during calculation and round the final result to the precision justified by the task.
This is not a universal mechanical rule; some algorithms require careful precision management throughout. But for ordinary school calculations, premature rounding is a frequent source of unnecessary discrepancy.
58. Catastrophic Cancellation
When two nearly equal floating-point numbers are subtracted, leading digits can cancel and leave a result dominated by representation error.
This is called catastrophic cancellation. It shows why algebraically equivalent formulas may have very different numerical reliability.
Numerical approximation therefore cares about representation and algorithm design, not only mathematical equivalence.
59. Error Budgets
An error budget allocates acceptable uncertainty across components of a system.
If a final measurement must be accurate within a given tolerance, engineers can assign portions of that tolerance to sensors, calibration, numerical integration, environmental variation and model approximation.
This turns “be accurate” into an operational plan.
60. Approximation Quality Depends on Consequence
A 5% approximation may be excellent for an early cost estimate and unacceptable for a medication dosage. A one-minute schedule error may be irrelevant for a picnic and serious for a tightly coordinated transport interchange.
Approximation quality cannot be judged without knowing what the answer will be used for.
61. Approximation and Mathematical Proof
Approximation does not weaken proof. Proof can establish rigorous facts about approximations:
- an error is below a stated bound;
- an iterative method converges;
- a numerical scheme is stable under specified conditions;
- an approximation ratio is guaranteed;
- a Taylor remainder has a controlled size.
Proof and approximation therefore work together: approximation supplies a usable answer; proof can supply guarantees about how far that answer may be from the exact one.
Related: How Mathematical Proof Works.
62. Approximation and Mathematical Generalisation
Generalisation asks what structure survives across many cases. Approximation asks what structure can be simplified without losing too much of what matters.
The two processes often interact. A local approximation may reveal a general pattern. A general theorem may provide an error bound for an approximation.
See How Mathematical Generalisation Works.
63. Approximation and Numerical Analysis
Numerical analysis is the specialist branch that studies algorithms for approximating mathematical problems on computers, including stability, conditioning, convergence and computational error.
This article sits one level broader. It treats approximation as a mathematical habit spanning estimation, measurement, school reasoning, modelling and numerical computation. The specialist numerical owner remains How Mathematics Works | Numerical Analysis.
64. A Practical Approximation Workflow
- State the quantity or problem being approximated.
- Decide why exact treatment is unnecessary or unavailable.
- Set the required tolerance.
- Choose an approximation method appropriate to the structure.
- Identify assumptions and valid range.
- Calculate using sufficient internal precision.
- Estimate, bound or empirically test the error.
- Check sensitivity to uncertain inputs.
- Compare the result with a rough order-of-magnitude estimate.
- Report only meaningful precision.
- State where the approximation should not be trusted.
65. A Student Approximation Checklist
- Am I being asked for an exact answer or an estimate?
- What precision is required?
- Which values have already been rounded?
- Should I use decimal places or significant figures?
- What interval does a rounded measurement represent?
- Have I rounded too early?
- Does my final magnitude make sense?
- Should I use ≈ instead of =?
- Can I state an upper or lower bound?
- Is the approximation valid in this range?
66. A Teacher’s Diagnostic View
When a learner produces a poor approximation, identify the failure type.
- scale failure: the learner cannot estimate order of magnitude;
- precision failure: decimal places and significant figures are confused;
- bound failure: a rounded value is treated as exact;
- method failure: the chosen approximation does not fit the problem;
- domain failure: a local approximation is used too far from its valid region;
- rounding failure: intermediate values are truncated too aggressively;
- interpretation failure: the final digits are reported with false precision;
- verification failure: no rough estimate or error check is performed.
These are different mathematical weaknesses and should not all be repaired with more calculator practice.
67. Approximation and Artificial Intelligence
AI systems frequently generate approximate answers: numerical estimates, probabilistic predictions, compressed representations and heuristic solutions.
The same rules apply. The system should distinguish exact computation from estimation, avoid inventing precision, preserve units, state uncertainty when material, and verify high-consequence numerical results independently.
A fluent decimal is not evidence of numerical reliability.
68. Worked Example: A Fast Shopping Estimate
Suppose four items cost $12.90, $7.20, $18.60 and $4.80.
A quick estimate:
13 + 7 + 19 + 5 = 44.
The exact total is $43.50. The estimate is close enough for checking whether $50 cash is sufficient. There is no need to optimise the approximation to the cent before the exact addition is available.
69. Worked Example: Measurement Bounds
A rectangular table is measured as 1.8 m by 0.9 m, each correct to the nearest 0.1 m.
The measurement intervals are:
1.75 ≤ L < 1.85 0.85 ≤ W < 0.95.
The nominal area 1.62 m² is useful, but it is not an exact physical area. A bound-aware analysis uses the endpoint intervals to understand the range compatible with the measurements.
70. Worked Example: Choosing an Approximation Method
Suppose we need √2.
- For mental checking, 1.4 may be sufficient.
- For ordinary school calculation, 1.414 may be sufficient.
- For a high-precision numerical task, many more digits may be required.
- If a rigorous interval is required, use a method that supplies a bound.
- If rapid iterative convergence is valuable and conditions are favourable, Newton’s method may be attractive.
There is no universally correct number of digits. Precision is a contract with the task.
71. Frequently Asked Questions
Is an approximation a wrong answer?
No. An approximation is a deliberately inexact answer. It becomes poor mathematics only when the error is uncontrolled, the precision is misrepresented or an exact answer was required.
Does more decimal places always mean more accuracy?
No. Extra digits can be meaningless if the inputs, measurements or model are uncertain.
What is the difference between error and mistake?
In approximation, error can be an unavoidable or deliberately accepted difference between an approximate and exact value. A mistake is an incorrect operation, assumption or interpretation.
Why not always calculate exactly?
Exact solutions may not exist in a convenient form, may require excessive computation, or may offer more precision than the real data justify.
How do I know whether an approximation is good enough?
Compare its error or error bound with the tolerance required by the task and check that the method’s assumptions remain valid.
What is the difference between approximation and estimation?
Estimation is one form of approximation, often used to obtain a quick value from simplified inputs. Approximation is the broader concept and includes numerical algorithms, series, interpolation, discretisation and many other methods.
72. Further Reading
- How Mathematics Works | Numerical Analysis
- How Mathematical Modelling Works
- How Mathematical Proof Works
- How Mathematical Generalisation Works
- MIT OpenCourseWare
- National Institute of Standards and Technology
73. Final Principle
Mathematical approximation works by making inexactness explicit and manageable.
It replaces an exact or difficult object with something simpler. It records what was simplified. It measures or bounds what was lost. It checks whether the loss matters. It refines the result only as far as the task justifies.
The deepest lesson is that mathematical reliability does not require pretending every useful answer is exact. Reliability comes from knowing how approximate the answer is, why it is approximate, where it remains trustworthy, and whether that is enough for the decision in front of us.
Continue the Mathematics route: How Mathematics Works · How Mathematical Generalisation Works · How Mathematics Works | Numerical Analysis