1 Definition and basic concepts

Truncation error is the difference between an exact mathematical quantity and the result obtained after an approximation is cut off at a finite stage. In practice, it appears when an infinite process is replaced by a finite computation, such as stopping a series after a few terms or using a grid to represent a continuous function. It is a fundamental measure of how much information is lost when a numerical method simplifies an ideal calculation.

The concept is used throughout calculus and numerical analysis because many exact expressions cannot be evaluated directly in closed form. By studying truncation error, one can estimate how closely a numerical method matches the underlying mathematical model and how the error changes as the approximation is refined.

1.1 Meaning of truncation

Truncation refers to the act of cutting off part of a mathematical process. A power series may be truncated after a finite number of terms, a derivative may be approximated by a finite difference formula, or an integral may be replaced by a finite sum. The omitted remainder is what creates truncation error.

The term does not imply a mistake in computation; rather, it describes the intentional simplification used to make a problem manageable. The size of the resulting error depends on the method and on how much of the original process is left out.

1.2 Difference from other types of error

Truncation error is one of several kinds of error studied in numerical work. It is associated with the mathematical approximation itself, whereas other errors may come from machine arithmetic, model assumptions, or measurement uncertainty. Distinguishing among these sources is important when analyzing accuracy.

1.2.1 Round-off error

Round-off error arises because computers store numbers with finite precision. Even if a numerical method were mathematically exact, its implementation on a machine would still involve rounding at each arithmetic step. This type of error is separate from truncation error, which comes from limiting the mathematical formula rather than from how the numbers are stored.

1.2.2 Discretization error

Discretization error is closely related to truncation error and often overlaps with it in practical analysis. It occurs when a continuous object, such as a function or differential equation, is represented on a discrete set of points. In many contexts, the error caused by discretizing a problem is obtained by truncating an exact local representation, so the two terms are sometimes used in related but not identical ways.

1.3 Local and global error

Local truncation error describes the error introduced in a single step or at a single point of approximation. Global error refers to the accumulated discrepancy after many steps or over an entire computation. A method may have small local error but still produce noticeable global error if the method is applied repeatedly or if errors accumulate over time.

The relationship between local and global error depends on the method’s stability and on how the errors propagate. In many numerical schemes, reducing local truncation error also reduces global error, though not always at the same rate.

2 Sources of truncation error

Truncation error appears in many kinds of approximation schemes. Its source is usually the replacement of an exact infinite or continuous process with a finite one that is easier to compute.

2.1 Series truncation

Infinite series provide exact representations for many functions, but they must be cut off after finitely many terms in actual computation. The omitted tail of the series is the truncation error.

2.1.1 Taylor series

When a function is approximated by a Taylor polynomial, only the first few terms of the Taylor series are retained. The remainder term measures the truncation error and depends on the size of the neglected higher-order derivatives. Near the expansion point, this error is often small, but it grows as one moves farther away.

2.1.2 Fourier series

Fourier series represent periodic functions as sums of sine and cosine terms. Using only finitely many harmonics produces a truncated Fourier series. The omitted modes create truncation error, which may appear as smoothing or oscillatory artifacts near discontinuities.

2.2 Finite-difference approximation

Finite-difference formulas approximate derivatives by using values of a function at nearby points. Because these formulas are derived from truncated Taylor expansions, they contain truncation error that depends on the spacing between sample points. Smaller spacing usually reduces the error, though not always without practical limits.

2.3 Numerical integration

Numerical integration replaces an exact integral by a finite weighted sum. Methods such as the trapezoidal rule and Simpson’s rule approximate the area under a curve using simple geometric or polynomial models. The difference between the exact integral and the numerical estimate is the truncation error of the quadrature formula.

2.4 Numerical methods for differential equations

In differential equations, truncation error arises when derivatives are approximated by discrete formulas or when continuous time is replaced by finite steps. Time-stepping methods for ordinary differential equations and finite-difference, finite-volume, or finite-element schemes for partial differential equations all involve some form of truncation. The resulting error reflects the mismatch between the exact differential operator and its discrete counterpart.

3 Error analysis

Error analysis studies how truncation error behaves as a method is refined. It provides a way to compare algorithms and to predict how accuracy changes with step size or with the number of retained terms.

3.1 Order of accuracy

The order of accuracy describes how quickly truncation error decreases as the approximation is improved. A method is said to be first-order, second-order, or higher-order depending on the leading power of the step size in its error term. Higher-order methods typically achieve better accuracy for the same resolution.

3.2 Asymptotic error terms

Asymptotic error terms express truncation error in simplified form for small step sizes or large numbers of retained terms. They show the dominant behavior of the error while ignoring smaller contributions. Such formulas are especially useful in determining which term controls the overall accuracy of a method.

3.3 Dependence on step size

In many numerical schemes, truncation error is tied directly to the step size or mesh spacing. Decreasing the step size often reduces the error, but the rate of improvement depends on the method’s structure.

3.3.1 Linear dependence

Some approximations produce errors proportional to the step size. In these cases, halving the step size roughly halves the leading truncation error. This behavior is common in lower-order formulas.

3.3.2 Higher-order dependence

Other methods have errors proportional to higher powers of the step size, such as the square or cube of the spacing. These methods converge more rapidly, so modest reductions in step size can lead to substantial improvements in accuracy.

4 Truncation error in calculus

Truncation error is especially visible in the numerical versions of standard calculus operations. Derivatives, integrals, and function expansions are all commonly approximated by finite formulas.

4.1 Approximating derivatives

Derivative approximations are often derived from local Taylor expansions. The resulting formula uses nearby function values, and the neglected terms determine the truncation error.

4.1.1 Forward difference

The forward difference uses the function value at a point and at a nearby point ahead of it. It is simple to implement but usually has a lower order of accuracy than centered formulas. Its truncation error is related to the next term in the Taylor expansion beyond the one used in the approximation.

4.1.2 Backward difference

The backward difference uses the current point and a point behind it. It is often useful when values are only available from past times or from one side of a boundary. Like the forward difference, its error comes from omitted higher-order terms.

4.1.3 Central difference

The central difference combines information from points on both sides of the target point. Because of symmetry, its truncation error is often smaller than that of one-sided formulas of comparable simplicity. This makes it a standard choice when balanced data are available.

4.2 Approximating integrals

Numerical integration formulas approximate an integral by summing contributions from subintervals or interpolation polynomials. Their truncation error measures the gap between the exact integral and the finite rule.

4.2.1 Riemann sums

Riemann sums divide an interval into finitely many subintervals and use sample values to estimate area. The approximation improves as the partition becomes finer, but the finite partition always leaves a truncation error unless the integrand has a very special form.

4.2.2 Newton–Cotes formulas

Newton–Cotes formulas approximate an integral by integrating an interpolating polynomial through equally spaced points. Examples include the trapezoidal rule and Simpson’s rule. Their truncation error depends on the degree of the polynomial used and on the smoothness of the integrand.

4.3 Approximating functions by finite expansions

Many functions are represented approximately by finite expansions such as Taylor polynomials or truncated orthogonal series. The truncation error is the difference between the original function and the finite approximation. This error can be analyzed through remainder formulas or norm estimates, depending on the application.

5 Applications

Truncation error is central in practical computation because it determines how trustworthy a numerical result is. It guides method selection, parameter choice, and refinement strategies.

5.1 Numerical differentiation

When derivatives are estimated from sampled data, truncation error limits the achievable precision. Engineers and scientists use error estimates to choose a formula that balances accuracy, data availability, and computational cost.

5.2 Numerical integration

In numerical integration, truncation error helps determine how many subintervals or evaluation points are needed. It is also used to compare quadrature rules and to decide whether a higher-order formula is worth the additional effort.

5.3 Solving ordinary differential equations

For ordinary differential equations, truncation error measures how closely a time-stepping method follows the exact solution over one step. Step size and method order strongly influence the long-term quality of the computed trajectory.

5.4 Solving partial differential equations

For partial differential equations, truncation error is used to assess the quality of discrete spatial and temporal approximations. It plays an important role in the design of finite-difference, finite-volume, and finite-element methods, where the goal is to represent a continuous system with a computable discrete model.

6 Reducing truncation error

Several strategies can reduce truncation error, though each has trade-offs involving cost, stability, or complexity.

6.1 Using smaller step sizes

A smaller step size generally improves the approximation by capturing more detail of the underlying function or solution. This is one of the most direct ways to reduce truncation error, although the benefit may be offset by increased computation or by round-off effects.

6.2 Using higher-order methods

Higher-order methods include more terms in the local approximation and therefore leave smaller leading error terms. They often provide better accuracy than low-order methods for the same number of evaluation points, especially when the solution is smooth.

6.3 Richardson extrapolation

Richardson extrapolation combines results computed at different step sizes to cancel leading truncation error terms. It can significantly improve accuracy when the error behaves in a predictable power-law manner. The technique is widely used in numerical analysis to refine estimates without redesigning the entire algorithm.

6.4 Adaptive methods

Adaptive methods adjust step size, polynomial degree, or mesh resolution according to the estimated local error. By concentrating effort where the function changes rapidly, they reduce truncation error efficiently and avoid unnecessary work in smoother regions.

7 Examples

Concrete examples help show how truncation error appears in familiar formulas.

7.1 Taylor polynomial remainder

If a smooth function is approximated by a Taylor polynomial of low degree, the remainder term gives the truncation error. For example, using a quadratic approximation near a point captures the function’s value, slope, and curvature, while ignoring higher derivatives. The error becomes smaller as the evaluation point approaches the center of expansion.

7.2 Finite difference derivative error

A derivative estimated from two nearby sample values is not exact because the formula omits higher-order Taylor terms. The leading truncation error typically involves a derivative of higher order multiplied by a power of the step size. This explains why closer sampling usually improves the estimate.

7.3 Quadrature error example

When an integral is approximated by a rule such as the trapezoidal rule, the exact curve is replaced by straight-line segments. The difference between the curved region and the segmented approximation produces truncation error. For smooth integrands, the error is often small and decreases as the interval is subdivided.

Truncation error is part of a broader framework for understanding numerical methods. It interacts with several other core ideas in approximation theory and computation.

8.1 Consistency and convergence

A method is consistent if its truncation error tends to zero as the discretization is refined. Convergence means that the numerical solution approaches the exact solution under refinement. Consistency is usually necessary for convergence, though not sufficient by itself.

8.2 Stability

Stability concerns how a method responds to small perturbations. Even a scheme with small truncation error can perform poorly if errors grow rapidly during computation. Stability helps determine whether local inaccuracies remain controlled or become amplified.

8.3 Error propagation

Error propagation describes how local errors influence later stages of a computation. Truncation error may accumulate, interact with round-off error, or be damped depending on the algorithm. Understanding propagation is essential for predicting the final accuracy of a numerical result.