What a Support Formula Tells You
A probability formula often includes a condition such as \(x>a\), \(0<x<1\), or \(x\in\{0,1,2,\ldots\}\). That condition gives important information about the values a random variable can take, but reading it correctly requires care. In particular, a density can be zero at a boundary point that still belongs to the support, and a discrete distribution can have support points with zero point probability.
The support describes where the distribution is located, including limit points that may not themselves receive probability. Parameter values can move a support boundary, leave it fixed, or make a formula invalid. The first step is to define support in terms of probabilities, rather than relying only on the displayed value of a density or mass function.
This definition works for discrete, continuous, and mixed distributions. For a discrete distribution, every point with positive point probability is in the support, but the support can contain additional limit points. For a density, it is the probability in neighborhoods that matters: the value assigned to the density at one individual point does not determine whether that point is in the support.
Basic Properties of Support
Two properties justify common ways of reading support formulas. First, support is closed. Second, it has probability one. These facts explain why boundaries and limit points are included even when they carry no probability individually.
Proof. If \(x\) is outside the support, then by definition some open interval around \(x\) has probability zero. Every point sufficiently close to \(x\) also has a smaller open neighborhood contained in that zero-probability interval. Those nearby points are outside the support too. Hence the complement of the support is open, so the support is closed.
To see that the support has probability one, take any \(x\) outside it. There is an open interval around \(x\) with probability zero. Within that interval we can choose an open interval with rational endpoints that still contains \(x\). There are only countably many rational-endpoint intervals, so the complement of the support is contained in the union of those rational-endpoint intervals that have probability zero. This is a countable union of probability-zero sets and therefore has probability zero. Thus \(P(X\in\operatorname{supp}(X))=1\). \(\square\)
If \(X\) has a density \(f\), the probability of an interval is the integral of \(f\) over that interval. Consequently, $$ \operatorname{supp}(X)=\left\{x:\int_{x-\varepsilon}^{x+\varepsilon} f(t)\,dt>0\text{ for every }\varepsilon>0\right\}. $$ The formula is unaffected by changing \(f\) at individual points, since such changes do not alter any of these integrals.
Justification. If \(x\in[a,b]\), every open interval around \(x\) intersects \((a,b)\) in an interval of positive length. The density is positive almost everywhere there, so its integral over that intersection is positive. Thus every neighborhood of \(x\) has positive probability. If \(x<a\) or \(x>b\), there is a sufficiently small neighborhood disjoint from \([a,b]\), and the density is zero almost everywhere on that neighborhood. Its probability is zero, so \(x\) is outside the support. This proves the stated support. \(\square\)
A Practical Reading Method
When a probability formula depends on a parameter, separate three questions: what parameter values make the formula a valid probability law, which values of \(x\) have positive probability in neighborhoods, and which boundaries are included in the support. A displayed condition on \(x\) is a strong clue, but the neighborhood definition settles boundary cases.
Confirm that the stated parameter range makes the mass or density nonnegative and normalized. Do not use a support formula at parameter values for which it is not a probability law.
Identify where the formula assigns positive mass or is positive on a region of positive length. For a density, a single point where the formula is zero or undefined does not settle support membership.
Include boundary and limit points when every neighborhood has positive probability. For countably supported distributions, the support is the closure of the set of values with positive point probability.
Ask whether changing a parameter moves the region of possible values or only changes probabilities within that region.
Worked Examples
Worked Example: A Uniform Density with Moving Boundaries
For \(a<b\), consider the density $$ f(x;a,b)=\frac{1}{b-a}\mathbf{1}\{a<x<b\}. $$ Find its support and state the parameter restriction.
The parameter restriction is \(a<b\), which makes the interval length \(b-a\) positive. The density is nonnegative, and its integral is $$ \int_{-\infty}^{\infty} f(x;a,b)\,dx =\int_a^b \frac{1}{b-a}\,dx =\frac{b-a}{b-a}=1. $$ So it is a probability density for every \(a<b\).
Every point strictly between \(a\) and \(b\) has a neighborhood with positive density and hence positive probability. A neighborhood of \(a\), no matter how small, contains an interval of positive length to the right of \(a\) within \((a,b)\); its probability is positive. The same reasoning applies at \(b\). Points outside \([a,b]\) have a small neighborhood disjoint from \((a,b)\). Therefore the support is \([a,b]\), although the formula uses the strict inequalities \(a<x<b\). Changing \(a\) or \(b\) moves a support boundary.
Worked Example: A Parameter That Changes Shape but Not Support
For \(\theta>-1\), let $$ f(x;\theta)=(\theta+1)x^\theta\mathbf{1}\{0<x<1\}. $$ Verify the parameter restriction and determine the support.
For \(0<x<1\), \(x^\theta\) is positive for every real \(\theta\). The integral is finite precisely when \(\theta>-1\), and under this condition $$ \int_{-\infty}^{\infty} f(x;\theta)\,dx =(\theta+1)\int_0^1 x^\theta\,dx =(\theta+1)\left[\frac{x^{\theta+1}}{\theta+1}\right]_0^1 =1. $$ Thus \(\theta>-1\) makes the formula a density.
The density is positive throughout \((0,1)\), whatever the allowed value of \(\theta\). Every neighborhood of either endpoint has positive probability, while points outside \([0,1]\) have a neighborhood of probability zero. Hence the support is \([0,1]\) for every \(\theta>-1\). The parameter changes how probability is distributed inside that interval, not the support boundaries.
Worked Example: A Shifted Exponential Density
Let \(\lambda>0\) and \(\mu\in\mathbb{R}\), and suppose $$ f(x;\mu,\lambda)=\lambda e^{-\lambda(x-\mu)}\mathbf{1}\{x>\mu\}. $$ Check the normalization and determine which parameter controls the support boundary.
Substitute \(u=x-\mu\). Then $$ \int_{-\infty}^{\infty}f(x;\mu,\lambda)\,dx =\int_\mu^\infty \lambda e^{-\lambda(x-\mu)}\,dx =\int_0^\infty \lambda e^{-\lambda u}\,du =\left[-e^{-\lambda u}\right]_0^\infty=1, $$ where \(\lambda>0\) ensures \(e^{-\lambda u}\) tends to zero as \(u\) tends to infinity. The density is positive for \(x>\mu\), so every neighborhood of \(\mu\) has positive probability, even though the density formula assigns zero at \(x=\mu\). Points below \(\mu\) have a neighborhood of probability zero. Thus the support is \([\mu,\infty)\). The parameter \(\mu\) moves the lower boundary; changing \(\lambda\) does not change it.
Worked Example: A Discrete Support with a Zero-Mass Limit Point
Suppose \(P(X=1/n)=2^{-n}\) for \(n=1,2,\ldots\), and \(P(X=x)=0\) for all other \(x\). Find the support and check whether zero belongs to it.
The probabilities sum to one: $$ \sum_{n=1}^{\infty}2^{-n}=1. $$ Each \(1/n\) is in the support because every neighborhood of \(1/n\) contains that point, which has positive probability. Also, every neighborhood of zero contains some \(1/n\) with positive probability: given \(\varepsilon>0\), choose \(n>1/\varepsilon\), so \(0<1/n<\varepsilon\). Therefore zero is in the support even though \(P(X=0)=0\).
If \(x<0\), a small neighborhood of \(x\) contains none of the values \(1/n\), so \(x\) is outside the support. If \(x>0\) is not one of the values \(1/n\), it is not a limit point of that sequence; it has a small neighborhood containing none of the values, so it is outside the support. The support is consequently \(\{0\}\cup\{1/n:n=1,2,\ldots\}\): the closure of the positive-mass values, not just those values themselves.
Transformations and Support Boundaries
A simple transformation rule is useful when a formula describes a shifted or rescaled random variable. For real constants \(a\) and \(b\ne 0\), let \(Y=a+bX\). Then $$ \operatorname{supp}(Y)=\{a+bx:x\in\operatorname{supp}(X)\}. $$
To verify this, write \(y=a+bx\). A neighborhood \((y-\varepsilon,y+\varepsilon)\) for \(Y\) corresponds under the one-to-one map \(u\mapsto a+bu\) to a neighborhood of \(x\) for \(X\), with radius \(\varepsilon/|b|\). The two neighborhoods have positive probability together or zero probability together. Thus every neighborhood of \(y\) has positive probability exactly when every neighborhood of \(x\) does. This proves the rule. If \(b=0\), then \(Y=a\) with probability one and its support is \(\{a\}\).
Worked Example: Reversing Endpoints Under a Negative Scale
Suppose \(X\) has support \([0,1]\) and \(Y=3-2X\). Determine the support of \(Y\).
Apply the transformation rule to the two endpoints: \(x=0\) maps to \(y=3\), and \(x=1\) maps to \(y=1\). Since the transformation reverses order, the resulting interval is \([1,3]\). In set notation, $$ \operatorname{supp}(Y)=\{3-2x:0\le x\le 1\}=[1,3]. $$ The endpoints remain included because the affine map is continuous and one-to-one.
Common Mistakes and Why Support Matters
- Equating support with points of positive mass. That misses limit points. The support of the distribution concentrated at \(1/n\) with probabilities \(2^{-n}\) includes zero, even though zero has no point mass. For a countably supported distribution, take the closure of the positive-mass values.
- Reading a strict density inequality as an open support. A density defined for \(a<x<b\) can have support \([a,b]\), because every neighborhood of either endpoint has positive probability.
- Using the density's value at one point. A density value at an individual point can be changed without changing the distribution. Check neighborhood probabilities instead.
- Ignoring parameter constraints. A formula that integrates to one only for \(\theta>-1\), for example, cannot be treated as a probability density outside that range.
- Assuming every parameter changes support. A parameter may alter probabilities within a fixed interval while leaving its endpoints unchanged; check the formula’s allowed region for \(x\).
Support is useful for checking whether proposed values are possible under a model, identifying parameter-dependent boundaries, and interpreting transformations. It also provides a disciplined way to read formulas: first check that the parameters define a probability law, then use positive probability in every neighborhood to settle the support.
Check Your Understanding
Use the neighborhood definition of support in each answer.
- A density is positive almost everywhere on \((2,5)\) and zero almost everywhere outside \([2,5]\). What is its support, and why are the endpoints included?
- For the density proportional to \(x^\theta\) on \(0<x<1\), what parameter restriction makes it integrable, and does that parameter change the support?
- A discrete random variable has positive mass at each \(1/n\), for \(n\ge1\), and no mass elsewhere. Does zero belong to its support? Explain.
- If \(X\) has support \([-1,2]\) and \(Y=4+3X\), find the support of \(Y\).
- Why is the set where a density is strictly positive not, by itself, always the support?