Interval arithmetic: verified numerical computation and error bounding

Basic operations and properties

Elementary operations are defined so the result covers all combinations of endpoints. For example, addition is [a,b]+[c,d] = [a+c, b+d]; subtraction is [a,b]-[c,d] = [a-d, b-c]. Multiplication and division require examining multiple endpoint products and, for division, additional care when the divisor interval contains zero. Implementations typically use directed rounding (outward rounding) so that floating-point endpoints are adjusted to guarantee containment despite rounding. A simple demonstration appears in many introductions: [1,2] + [3,4] = [4,6]. Interval functions are built from these operations or by taking the range of a function over an interval; one standard construction is the natural interval extension, which substitutes interval arguments into an expression and evaluates with interval arithmetic.

Representations and extensions

Interval arithmetic comes in several flavors: classical real intervals, affine forms and modal intervals are examples of richer representations that attempt to reduce overestimation and capture dependency between quantities. Interval matrices and interval systems of linear equations extend the idea to linear algebra. Software libraries implement these concepts for many programming languages; they typically provide basic interval types, elementary functions, and often algorithms for root finding and global optimization. For introductions to libraries and implementation strategies, consult interval documentation and practical guides such as error analysis tutorials.

Applications and examples

Interval arithmetic is widely used where guaranteed bounds are required. Typical applications include validated numerics (producing mathematically rigorous error bounds for computed quantities), propagation of measurement uncertainty (each measurement interval captures instrument error), robust control and verification of numerical algorithms, and global optimization where intervals help enclose minima or roots. Simple use-cases include computing bounds on a function given uncertain inputs, or verifying that no root exists in a region. For practical case studies and sample code, see worked examples and implementation notes at practical guides.

Limitations and notable facts

Despite its strengths, interval arithmetic has limitations. The dependency problem leads to overestimation: evaluating the same variable multiple times in an expression can produce unnecessarily wide intervals. For instance, x-x evaluated with x=[1,2] yields [1-2,2-1] = [-1,1] instead of the exact {0}. Specialized techniques—such as algebraic reformulation, domain splitting, or using affine arithmetic—help to mitigate this issue. Another practical concern is performance: guaranteed containment requires conservative rounding and sometimes additional computations, which can be slower than conventional floating-point arithmetic.

History and practical considerations

The modern systematic study of interval methods dates to mid-20th century numerical analysis, with further formalization in the work of researchers like R. E. Moore and others who developed the theoretical foundations and algorithms. Today interval techniques form a core tool in validated computing and appear in many scientific and engineering toolkits. When adopting interval arithmetic, practitioners must balance the need for rigorous bounds against increased computational cost and potential over-conservatism. For further reading and authoritative surveys, consult accessible overviews at further reading.

• Key idea: represent uncertainty by intervals and propagate them through computations to maintain proven bounds.
• Practical tip: transform expressions and split domains to reduce overestimation.
• Caveat: division by intervals containing zero is undefined and must be treated specially.