Continuous Uniform Distribution¶

The bounded continuous probability law whose density is constant, assigning probability in direct proportion to interval length.

Version: v2 · 2026-09-06 · History
Domain-specific #: 1554
Origin domain: probability
Subdomain: continuous distributions
Aliases: Rectangular Distribution, Uniform Distribution on an Interval

Core Idea¶

The continuous uniform distribution on a finite interval assigns the same probability density to every point of that interval. If \(a<b\), a random variable \(X\sim U(a,b)\) has density

\[ f_X(x)= \begin{cases} \dfrac{1}{b-a}, & a\le x\le b,\\ 0, & \text{otherwise}. \end{cases} \]

Consequently, for any subinterval \([c,d]\subseteq[a,b]\),

\[ \Pr(c\le X\le d)=\frac{d-c}{b-a}. \]

Equal length, not equal listed points, is the invariant. Each individual point still has probability zero. NIST describes the law as equal probability over a given continuous range and identifies \(U(0,1)\) as its standard form.^[1]

The abstraction is a specific probability-distribution family, not the substrate-neutral idea of fairness or sameness. It is reusable because its support, constant density, affine transformations, moments, sampling method, and hard-boundary implications recur across probability, simulation, measurement-error models, and reference calculations.

Structural Signature¶

Mandatory roles:

Finite ordered endpoints: real numbers \(a<b\) define the support.
Reference measure: ordinary interval length, formally Lebesgue measure on the real line.
Constant density: \(1/(b-a)\) throughout the support.
Zero exterior density: values outside \([a,b]\) are impossible under the model.
Length proportionality: subinterval probability equals its length divided by total support length.
Continuous random variable: singletons have probability zero despite equal density.
Location-scale transformation: \(a+(b-a)U\) maps \(U\sim U(0,1)\) to \(U(a,b)\).

Invariants: normalization integrates to one; translation changes both endpoints without changing shape; positive rescaling preserves uniformity; and midpoint symmetry makes mean and median equal to \((a+b)/2\).

Recognition test. Identify the interval and check density constancy relative to length. A finite list of equally likely outcomes is discrete uniform, not continuous uniform.

What It Is Not¶

It is not the discrete uniform distribution, which assigns positive mass to each member of a finite or countable support. It is not a normal distribution with a large variance: the normal has unbounded tails and a nonconstant density.

It is not “complete ignorance” without qualifications. Uniformity depends on a coordinate and reference measure. A parameter uniform in \(x\) is generally not uniform after a nonlinear reparameterization \(y=g(x)\). It is also not an improper uniform prior on the entire real line; no positive constant density over an infinite-length line can integrate to one.

It is not a claim that observed samples will be evenly spaced. Independent uniform draws cluster and leave gaps by chance. Nor is every bounded variable reasonably modeled as uniform: empirical concentration near a center or boundaries contradicts the constant-density assumption.

Scope of Application¶

Continuous uniform laws model quantities known only to lie in a bounded interval when equal-length regions are assigned equal probability. Examples include an idealized random phase on a cycle represented over one period, randomized start times within a window, simulation inputs sampled over fixed limits, and rounding error under specific phase assumptions.

The standard uniform distribution is central to simulation. Pseudorandom generators typically expose values approximating \(U(0,1)\); transformations then generate other laws. NIST highlights random-number generation as a major application of the standard form.^[1]

In statistical inference, \(U(a,b)\) is also a nonregular family when endpoints are unknown because the support depends on parameters. Sample minimum and maximum then carry direct endpoint information, and familiar likelihood regularity shortcuts may fail.^[2] The entry does not treat every inference theorem for this model, but it preserves the support-dependence boundary.

Uniformity extends to geometric regions through normalized area or volume, yet those are uniform distributions with respect to a stated measure, not automatically the one-dimensional interval family. This node centers the real bounded interval.

Clarity¶

“Every value is equally likely” is useful intuition but mathematically misleading for a continuum. Any exact value has probability zero. The precise statement is that equal-length measurable subsets receive equal probability, and probability is the integral of a constant density.

Endpoint inclusion does not change the distribution: \([a,b]\), \((a,b)\), and half-open variants differ only by sets of probability zero. It may matter for software return conventions, but not for the probability law.

Two parameterizations recur. Endpoint form uses \(a,b\). Location-scale form uses a center \(m=(a+b)/2\) and width \(w=b-a\). Calculations should state which is used. The word “rectangular” describes the density plot, not the cumulative distribution, whose interior segment is linear.

Manages Complexity¶

The law reduces an uncertain bounded quantity to two parameters. Once \(a\) and \(b\) are fixed, normalization, interval probabilities, quantiles, moments, and sampling all follow immediately. NIST gives mean \((a+b)/2\) and standard deviation \((b-a)/\sqrt{12}\).^[1]

The quantile function is especially simple:

\[ F^{-1}(p)=a+p(b-a),\qquad 0\le p\le1. \]

This makes the family a base interface for Monte Carlo construction and a transparent benchmark against which nonuniform assumptions can be compared.

Its simplicity also exposes modeling risk. All interior locations are exchangeable under translation within the interval, but boundaries are absolute discontinuities in possibility. If the real process can exceed them, the model assigns those observations probability zero rather than merely low probability.

Abstract Reasoning¶

Normalization follows because the rectangular area is \((b-a)\times1/(b-a)=1\). Integrating \(x/(b-a)\) over the interval yields

\[ \mathbb E[X]=\frac{a+b}{2}, \qquad \operatorname{Var}(X)=\frac{(b-a)^2}{12}. \]

For \(U\sim U(0,1)\), the event \(a+(b-a)U\le x\) is equivalent to \(U\le(x-a)/(b-a)\), giving the piecewise-linear cumulative distribution. Conversely, standardization \((X-a)/(b-a)\) produces \(U(0,1)\).

If \(Y=g(X)\) for a differentiable monotone nonlinear function, the transformed density includes the Jacobian factor \(|d g^{-1}(y)/dy|\). Therefore uniformity is not generally invariant under nonlinear coordinate changes. This is the key diagnostic against interpreting uniformity as coordinate-free ignorance.

Knowledge Transfer¶

Literal transfer occurs whenever the same bounded interval, constant density, and length measure are present. Simulation, reliability windows, randomized algorithms, and simplified physical phase models reuse the exact equations.

Transfer to a disk or sphere requires replacing interval length with normalized area or surface measure. The higher-level pattern—constant density relative to a chosen base measure—survives, but the one-dimensional formulas do not. Transfer to discrete cases replaces density with point mass and therefore changes the identity.

The catalog parent Probability Distribution supplies the general mapping from events to probabilities. Continuous Uniform adds the exact support geometry and constant-density rule.

Examples¶

Random time in a window. If an event time is sampled uniformly between 2 p.m. and 3 p.m., the probability it falls in the first 15 minutes is \(15/60=1/4\). The probability of exactly 2:17:00.000… is zero in the continuous idealization.

Standard simulation input. If \(U=0.30\) is a standard-uniform draw and the target interval is \([10,20]\), then \(X=10+10U=13\). Repeating with independent standard draws simulates \(U(10,20)\), subject to the generator's finite-precision approximation.

Failed fit. A manufacturing tolerance of \([9.9,10.1]\) does not make measured diameters uniform. If production is centered around 10 with declining tails, the bounds describe acceptance, not the data-generating density.

Structural Tensions¶

Equal density versus zero point probability: equality concerns infinitesimal density and lengths, not positive mass at each point. Diagnostic: ask for the probability of a singleton; a positive answer signals a discrete model.
Simplicity versus hard boundaries: two parameters make reasoning easy but assign zero probability immediately outside the support. Diagnostic: test whether observations beyond \(a\) or \(b\) are genuinely impossible or merely unlikely.
Uniformity versus coordinate choice: constancy in one parameterization need not survive nonlinear transformation. Diagnostic: compute the Jacobian under the proposed reparameterization.
Random sampling versus visual evenness: independent uniform samples can cluster. Diagnostic: distinguish distributional equality over repeated draws from spacing constraints within one sample.
Autonomy versus generic probability distribution: the parent supplies probability axioms but not bounded constant density. Diagnostic: subtract generic event weights; if interval support, length proportionality, and affine standardization remain, the family is autonomous.

Structural–Framed Character¶

The distribution is formal and structural: interval, measure, density, and transformation fully determine it. Its domain framing is mathematical rather than institutional. No normative claim that uniform treatment is fair or desirable follows from the probability model.

The coordinate-measure dependence is part of its structural character, not a defect. A complete uniformity claim always names the support and the measure relative to which density is constant.

Structural Core vs. Domain Accent¶

The portable core is equal allocation relative to equal measure on a bounded support. The domain accent consists of real-valued random variables, Lebesgue density, integration, and statistical parameterization.

Removing those mathematical commitments yields generic evenness, already too broad to identify this distribution. Keeping them produces exact recognition across probability and statistics without supporting prime promotion.

domain_specific:probability_distribution is the minimal parent because \(U(a,b)\) is a particular continuous probability law. Random Variable is related as a carrier of the law, not a superclass of the distribution itself. Distributional Assumption describes its modeling use but not its mathematical identity.

Relationships to Other Abstractions¶

Current abstraction Continuous Uniform Distribution Domain-specific

Parents (1) — more general patterns this builds on

Continuous Uniform Distribution is a kind of Probability Distribution Domain-specific

domain_specific:probability_distribution is the minimal parent because \(U(a,b)\) is a particular continuous probability law.

Neighborhood in Abstraction Space¶

Continuous Uniform Distribution sits in a sparse region of the domain-specific corpus (86^th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Unclustered & Miscellaneous (1565 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-09-08

Not to Be Confused With¶

Discrete uniform distribution: equal positive masses on discrete outcomes.
Improper uniform prior: a nonnormalizable constant over an unbounded set.
Uniform measure on another space: requires a different support and reference measure.
Normal distribution: unbounded and centrally concentrated.
Uniform random number generator: an implementation approximating samples, not the law itself.
Evenly spaced design: deterministic spacing rather than independent random draws.

References¶

[1] NIST/SEMATECH, “Uniform Distribution,” e-Handbook of Statistical Methods, §1.3.6.6.2. https://www.itl.nist.gov/div898/handbook/eda/section3/eda3662.htm registry ↩a ↩b ↩c

[2] George Casella and Roger L. Berger, Statistical Inference, 2^nd ed., Duxbury, 2002, treatments of continuous distributions, transformations, and endpoint-parameter families. registry ↩