Is a constant function simple?

Yes. It takes one value and is measurable on the usual measurable-domain setup.

Is a histogram related to simple functions?

A histogram's piecewise constant height function is typically a simple function when it uses finitely many measurable bins.

Can simple functions approximate a continuous function?

Yes. Suitable finite-level approximations can approach a function increasingly closely; the approximation scheme and convergence statement must be specified.

AI Glossary / Mathematics and optimization / Advanced

Simple function

Definition

A simple function is a measurable function that takes only finitely many distinct values. It can be expressed as a finite sum of constants multiplied by indicator functions of measurable sets. In integration theory, simple functions provide basic building blocks for more general functions.

What it means

The word simple has a precise mathematical meaning here. It does not mean easy to understand, computationally cheap or linear. A function can have a complicated domain partition and still be simple if its output range contains only finitely many values.

For example, a scoring rule that returns 0, 2 or 5 depending on which measurable region contains its input is a simple function. The domain may contain infinitely many possible inputs; only the number of distinct output values must be finite.

This concept connects naturally to binning, categorical scoring and finite-valued approximations. Its main mathematical role is foundational: integrals of nonnegative measurable functions can be developed through suitable simple-function approximations.

How it works

Write s(x)=Σⱼaⱼ1Aj(x), where 1Aj(x) is one when x belongs to Aj and zero otherwise. A canonical representation uses disjoint measurable sets, one for each output value, so exactly the appropriate coefficient contributes at each input.

With a measure μ, the integral of a nonnegative simple function over disjoint level sets is Σⱼaⱼμ(Aj). On a probability space, the corresponding expectation is a weighted average of its values, using the probabilities of the level sets.

Measurability matters because those sets must have meaningful measures. Finite range alone does not guarantee measurability on an arbitrarily specified measurable space. Signed simple functions can also be used, but finite integrals require appropriate treatment when the underlying measure is infinite.

An indicator function is the simplest example, with values zero and one. A piecewise constant rule with finitely many measurable pieces is simple. A function with infinitely many distinct outputs, such as x↦x on [0,1], is not simple under this definition, even though its formula is short.

Three practical examples

A probability instructor computes the expected value of a finite-valued scoring rule by weighting each score by the probability of its region.
A numerical analyst approximates a nonnegative function using successively finer finite output levels.
A data scientist relates a binned feature representation to a piecewise constant measurable function.

Think of it this way

Think of a map colored with a fixed palette. Many locations can share a color, and the boundaries can be complicated, but only finitely many colors appear.

A simple function assigns numerical values rather than colors, and its regions must be measurable for integration. The palette analogy does not mean that every measurable region has a visually tidy boundary.

Worked example

Integrating a three-level rule

On [0,1], define s(x)=2 for 0≤x<1/4, s(x)=5 for 1/4≤x<3/4, and s(x)=1 for 3/4≤x≤1. The three regions are disjoint and cover the domain.

Using ordinary interval length as the measure, their measures are 1/4, 1/2 and 1/4. The integral is 2×1/4+5×1/2+1×1/4=13/4=3.25. If X is uniform on [0,1], this is also E[s(X)].

The squared score has expectation 4×1/4+25×1/2+1×1/4=55/4=13.75. Its variance is 55/4−(13/4)²=51/16=3.1875.

If a different distribution assigns probabilities 0.1,0.6 and 0.3 to the same regions, the expected score becomes 2×0.1+5×0.6+1×0.3=3.5. The function has not changed; the measure used to weight its regions has.

Endpoint assignments do not affect the uniform calculation because individual points have zero probability. They can matter under a distribution that places positive probability on a boundary.

Illustrative case study

A fictional service-planning team assigns three workload scores to ranges of a normalized demand indicator. An analyst averages the three scores equally and treats that as the expected workload score.

A colleague points out that the ranges occur with different frequencies. The team computes a probability-weighted average and documents exactly which range contains each boundary value.

The scoring rule remains a finite-valued function, but the team avoids confusing its mathematical simplicity with empirical validity. Whether the three scores represent meaningful workload differences must be assessed against actual service observations.

Common misconceptions and mistakes

Simple means linear. A simple function can be nonlinear and discontinuous.
A finite domain is required. An infinite domain can map to a finite set of values.
Every piecewise constant function is simple. Infinitely many distinct levels would violate the finite-range requirement.
The arithmetic mean of the levels is always its expectation. Region probabilities supply the weights.
Boundary choices never matter. Atoms of probability at a boundary can make them consequential.

Business use

For most business readers, the practical connection is to finite scoring rules and discretized representations. Thinking in terms of level sets clarifies which inputs receive which score and how to aggregate those scores correctly.

Document the partition, boundary conventions and probability population used for any expectation. A rule may be easy to implement yet hide substantial variation within a bin. If it approximates a continuous quantity, quantify the approximation error where possible and investigate whether important decisions occur near boundaries.

Check your understanding

A function on all real numbers returns one of three scores, but the regions have probabilities 0.1,0.6 and 0.3. Should its expectation be the unweighted mean of the scores?

Answer: No. Multiply each score by its region's probability and add. An unweighted mean would assume equal probabilities that were not given.

Related terms

Indicator function
Lebesgue integral
Measurable function
measure
expectation
binning

Learn next

Read indicator function and expectation. On [0,1], form s(x)=floor(4x)/4 and list its possible values, including x=1. Check that it is simple and that 0≤x−s(x)<1/4 throughout the interval.

Indicator function

Sources

Aalto Dictionary of Machine Learning, simple function: https://aaltodictionaryofml.github.io/ADictML_Main.pdf. Source for finite-valued measurability and indicator representation. The three-level integration, variance calculation and fictional workload scenario are original teaching illustrations.

Previous: Simple effect Next: Simple interaction