Normalizing a Joint Density

Why Normalization Matters

A formula can look like a plausible joint density and still fail to describe any probability model. To be a valid density, it must assign no negative probability locally, and its total probability across the entire support must equal one. Checking these conditions is the first step before using a joint density to calculate probabilities.

In many models, the proposed function contains an unknown constant. Integrating the function over its support gives an equation for that constant. The calculation is straightforward only if the support and the limits of integration are read correctly: the integral is over the region where the density is defined, not automatically over all possible values of the random variables.

Definition: A joint probability density function \(f_{X,Y}(x,y)\) for continuous random variables \(X\) and \(Y\) must satisfy two conditions: \(f_{X,Y}(x,y)\geq 0\) throughout its support, and its integral over the full support must equal one. For a stated support \(R\), the total probability condition is

\iint_R f_{X,Y}(x,y)\,dx\,dy=1.

Outside the support, the density is zero. Values assigned on boundary curves or lines do not affect these integrals.

Normalizing Over a Rectangular Support

For a rectangular support \(a<x<b,\ c<y<d\), the total integral can be computed as an iterated integral. If \(f_{X,Y}(x,y)=k g(x,y)\), where \(g\) is specified and \(k\) is an unknown constant, then

1=\int_a^b\int_c^d k g(x,y)\,dy\,dx =k\int_a^b\int_c^d g(x,y)\,dy\,dx.

If the integral of \(g\) over the support is a finite positive number \(I\), the required constant is \(k=1/I\). This value makes the total integral one. It does not, by itself, establish that the function is a density: the resulting function must also be nonnegative everywhere on the support.

The order of integration may be reversed on a rectangle when the integrals exist, but the limits must still describe exactly the same rectangle. In each example below, the limits are written explicitly so that both the support and the integration are visible.

Worked Example: Normalizing a Linear Density on a Rectangle

A hypothetical model assigns a proposed joint density proportional to \(x+y\) on the support \(0<x<2,\ 0<y<1\). Find the constant \(c\) that makes it a valid density, and check both density conditions.

Method. Integrate the proposed function over the full stated rectangle, set the result equal to one, and then check nonnegativity. The common mistake to avoid is integrating over a unit square rather than the rectangle whose \(x\)-width is two.

Write \(f_{X,Y}(x,y)=c(x+y)\) on the support. First calculate the integral without \(c\):

\begin{aligned} \int_0^2\int_0^1(x+y)\,dy\,dx &=\int_0^2\left[xy+\frac{y^2}{2}\right]_{y=0}^{y=1}\,dx\\ &=\int_0^2\left(x+\frac12\right)\,dx\\ &=\left[\frac{x^2}{2}+\frac{x}{2}\right]_{x=0}^{x=2}\\ &=(2+1)-0=3. \end{aligned}

The normalization equation is \(3c=1\), so \(c=\frac13\). The proposed density is therefore \(f_{X,Y}(x,y)=\frac13(x+y)\) on the support and zero outside it.

Since \(x>0\) and \(y>0\) on the support, \(x+y>0\), and multiplication by \(\frac13\) preserves nonnegativity. Its total integral is \(\frac13(3)=1\). Both conditions hold, so this is a valid joint density.

Check. The constant is positive, the function is nonnegative on the specified rectangle, and the total probability is exactly one.

Why the Integral Must Equal One

A joint density gives probability through integration: the probability that \((X,Y)\) lies in a region is the integral of the density over that region. The full support contains all possible values of the pair, so the event that \((X,Y)\) lies somewhere in the support has probability one. Applying the density rule to that event gives the normalization condition.

P\big((X,Y)\in R\big)=\iint_R f_{X,Y}(x,y)\,dx\,dy=1.

This also explains why checking only a few point values is not enough. Nonnegativity is a condition throughout the support, while normalization concerns the accumulated mass across the entire support. One condition cannot replace the other.

Worked Example: A Product Function With Different Side Lengths

A hypothetical joint model has \(f_{X,Y}(x,y)=kxy\) for \(0<x<1,\ 0<y<3\), and zero elsewhere. Find \(k\), then verify that the resulting function is a density.

Method. Integrate the product over the rectangle. The common mistake to avoid is treating the support as a unit square and omitting the factor contributed by the \(y\)-range from zero to three.

Separate the two factors by iterated integration:

\begin{aligned} \int_0^1\int_0^3 xy\,dy\,dx &=\int_0^1 x\left[\frac{y^2}{2}\right]_{y=0}^{y=3}\,dx\\ &=\int_0^1 x\left(\frac{9}{2}-0\right)\,dx\\ &=\frac{9}{2}\left[\frac{x^2}{2}\right]_{x=0}^{x=1}\\ &=\frac{9}{2}\cdot\frac12=\frac94. \end{aligned}

The total integral of \(kxy\) is \(k\cdot\frac94\). Set it equal to one:

k\cdot\frac94=1 \quad\Longrightarrow\quad k=\frac49.

On the support, \(x>0\), \(y>0\), and \(k=\frac49>0\). Thus \(kxy\geq0\). Its total integral is \(\frac49\cdot\frac94=1\), so both conditions hold.

Check. The longer \(y\)-interval contributes \(\int_0^3 y\,dy=\frac92\), and the \(x\)-interval contributes \(\int_0^1 x\,dx=\frac12\). Their product is \(\frac94\), as used in the normalization.

Nonnegativity and Normalization Are Separate Checks

A proposed function may integrate to one and still fail to be a density if it is negative somewhere on the support. Conversely, a nonnegative function is not a density unless its integral is one. When the formula includes an unknown constant, first solve the normalization equation and then examine the sign of the resulting function on the entire support.

For example, on the unit square the function \(1-2x\) takes negative values when \(x>\frac12\). Its integral over the square is zero, not one, so it fails both checks. More generally, an integral calculation alone cannot establish nonnegativity. Inspect the factors and the range of each variable, including endpoints if the support includes them.

Worked Example: Testing a Proposed Density With a Fixed Constant

A proposed model specifies \(f_{X,Y}(x,y)=\frac12(x+y)\) for \(0<x<1,\ 0<y<1\), and zero elsewhere. Determine whether this is a valid density. If it is not, state whether changing the constant could normalize the same functional form.

Method. First integrate the stated function to test its total mass, then check its sign. The common mistake to avoid is declaring a function valid merely because it is nonnegative.

The integral of \(x+y\) over the unit square is

\begin{aligned} \int_0^1\int_0^1(x+y)\,dy\,dx &=\int_0^1\left(x+\frac12\right)\,dx\\ &=\left[\frac{x^2}{2}+\frac{x}{2}\right]_{x=0}^{x=1}\\ &=\frac12+\frac12=1. \end{aligned}

Therefore the stated function has total integral \(\frac12\cdot1=\frac12\), not one. It is nonnegative on the unit square, but it is not a valid density because it has only half the required total mass.

For the same functional form \(c(x+y)\), normalization requires \(c\cdot1=1\), so \(c=1\). With this replacement, the function is nonnegative and integrates to one. The given function with coefficient \(\frac12\) is invalid; the functional form can be normalized by using coefficient \(1\).

Check. The original total mass is \(\frac12\), while the normalized version has total mass \(1\). The sign condition holds for both, but only the normalized version meets both requirements.

A Reliable Checking Procedure

For a proposed joint density with a stated support, use the same sequence each time. Keeping the support visible in the calculation helps prevent errors in both the integral and the final validity decision.

Record the support.
Write the full range of \(x\) and \(y\), and set the density equal to zero outside that region.

Check the sign.
Determine whether the proposed expression is nonnegative throughout the support. If an unknown constant is present, its required sign may depend on the sign of the remaining expression.

Integrate over the support.
Use limits that cover the entire support exactly once. Keep any unknown multiplicative constant outside the integral when convenient.

Normalize and conclude.
Set the total integral equal to one and solve. Substitute the result into the function, then state explicitly whether nonnegativity and total mass both hold.

Common Exam Traps

Using the wrong area. A rectangle with side lengths two and one has area two, not one. For a constant function, the total integral is the constant multiplied by the area.
Forgetting one factor in an iterated integral. In a product such as \(xy\), integrate with respect to the inner variable using its own limits before integrating the remaining expression.
Checking only the total integral. A function that is negative on part of the support cannot be a density, even if its integral happens to equal one.
Checking only nonnegativity. A nonnegative function with total integral less than or greater than one is not normalized and therefore is not a density as stated.
Ignoring where the formula applies. The density is integrated over the stated support and is zero elsewhere. Do not extend the limits just because the algebraic formula can be evaluated outside the support.

The essential check is short but has two parts: the proposed joint function must be nonnegative on its support, and its integral over that support must be one. When an unknown constant is the only obstacle, the total-integral equation determines it; the sign check still has to be stated separately.

Check Your Understanding

For each problem, identify the support in the integral and check nonnegativity as well as total mass where appropriate.

A proposed density is \(c(2x+y)\) on \(0<x<1,\ 0<y<2\), and zero elsewhere. Find \(c\) and determine whether the normalized function is a density. (Answer: \(c=\frac14\); it is nonnegative and integrates to one.)
A proposed density is \(cx^2y\) on \(0<x<2,\ 0<y<1\), and zero elsewhere. Find the normalizing constant and verify the total integral. (Answer: \(c=\frac34\); the unnormalized integral is \(\frac43\), so the normalized integral is \(1\).)
A model specifies \(f_{X,Y}(x,y)=\frac12(x+y)\) on the unit square and zero elsewhere. Is it a valid density as written? (Answer: No; it is nonnegative but its total integral is \(\frac12\).)
Find the constant \(c\) for \(c(1+x)\) on \(0<x<1,\ 0<y<2\), with zero density elsewhere. Check both conditions. (Answer: \(c=\frac13\); the unnormalized integral is \(3\), and the resulting function is nonnegative and integrates to one.)
Find the constant \(c\) for \(c(1-y/2)\) on \(-1<x<1,\ 0<y<2\), with zero density elsewhere. (Answer: \(c=\frac12\); the unnormalized integral is \(2\), and the function is nonnegative on the support.)

What You'll Learn