Common Errors in Independent-Sum Problems

Why Independent-Sum Calculations Go Wrong

A sum can look familiar and still be mishandled. The most consequential errors are assuming independence from marginal distributions, multiplying densities instead of convolving them, using support bounds that ignore the joint model, and applying a closure rule with incompatible parameters. These are not merely notation slips: each can produce a plausible-looking but incorrect distribution.

The first question is whether the variables are independent. If they are not, their marginal distributions alone generally do not determine the distribution of their sum. If they are independent, convolution is the general method; a standard distributional closure rule is a shortcut only when its assumptions and parameterization match. The checks in this tutorial are aligned with the SOA Exam P (Probability) syllabus.

Definition: For random variables $X$ and $Y$, their sum is $W=X+Y$. Independence means that the joint distribution factors into the product of the marginal distributions. In the continuous case, when a joint density exists, this means $f_{X,Y}(x,y)=f_X(x)f_Y(y)$. Knowing $f_X$ and $f_Y$ without knowing their dependence does not in general determine the distribution of $W$.

Start With the Joint Model and the Support

The support of a random variable is the set of values it can take, or, for a continuous variable, the set where its density can be positive, including boundary points in the support's closure. If $X$ and $Y$ are independent and their densities are positive throughout intervals $[a,b]$ and $[c,d]$, respectively, then the support of $X+Y$ is $[a+c,b+d]$. Without independence or another specified joint-distribution condition, the component densities alone do not determine the support of the sum.

Under independence, the possible component values combine across the product of their supports. But dependence can restrict which pairs $(X,Y)$ occur together. Consequently, adding marginal support endpoints is not a safe shortcut unless the required joint-model conditions hold.

Worked Example: Identical Marginals, a Different Sum

Let $X\sim\operatorname{Uniform}(0,1)$, and define $Y=1-X$. Find the support of $W=X+Y$ and $P(W\leq 1)$. Then explain why treating $X$ and $Y$ as independent would give the wrong result. Method: use the specified dependence before applying any independent-sum formula. Mistake avoided: inferring independence from the marginal distributions.

First, $Y=1-X$ implies that $Y$ also has a uniform distribution on $[0,1]$. Thus both marginal densities are positive on $[0,1]$. However, the variables are dependent: once $X$ is known, $Y$ is fixed.

For every outcome,

$$ W=X+Y=X+(1-X)=1. $$

Therefore, the support of $W$ is the single point $\{1\}$, and

P(W\leq 1)=P(1\leq 1)=1.

For comparison, if the two uniform variables were independent, convolution would give a triangular density on $[0,2]$. In that different model, $P(X+Y\leq 1)$ would be the area of the triangle $\{(x,y):x\geq0,\ y\geq0,\ x+y\leq1\}$ inside the unit square, namely $1/2$. That is not the model in this problem. As a check, the actual sum has $E[W]=1$, consistent with $W=1$ with probability one.

Convolution Is Not Multiplication

For independent continuous variables with densities, multiplying $f_X(x)$ and $f_Y(y)$ gives the joint density at the pair $(x,y)$; it does not give the density of their sum at a single value. To obtain the density of $W=X+Y$, the contributions from all pairs satisfying $x+y=w$ must be combined. This is what convolution does.

Theorem: If $X$ and $Y$ are independent continuous random variables with densities $f_X$ and $f_Y$, then the density of $W=X+Y$ is $$ f_W(w)=\int_{-\infty}^{\infty} f_X(x)f_Y(w-x)\,dx. $$ For independent discrete random variables, the corresponding rule is $P(W=w)=\sum_x P(X=x)P(Y=w-x)$, summed over values for which both probabilities are defined.

Here is the reason for the continuous formula. The joint density under independence is $f_X(x)f_Y(y)$. For a fixed $x$, the event $w<W\leq w+h$ requires $w-x<Y\leq w+h-x$. Integrating the joint density over these pairs, dividing by $h$, and letting $h$ approach zero gives the integral above, wherever the resulting density exists. The formula adds contributions along the line $x+y=w$; a pointwise product of the two marginal densities does not.

Worked Example: Convolving Uniform Densities on Different Intervals

Suppose $X\sim\operatorname{Uniform}(0,2)$ and $Y\sim\operatorname{Uniform}(1,2)$, independently. Find the density of $W=X+Y$, then calculate $P(W\leq 2.5)$. Method: convolve the densities by finding the overlap of their valid intervals. Mistake avoided: multiplying the densities at a single pair and overlooking that the overlap changes with $w$.

The densities are $f_X(x)=1/2$ on $[0,2]$ and $f_Y(y)=1$ on $[1,2]$, and zero elsewhere. The independent support bounds give $1\leq W\leq4$. In the convolution integral, both densities are nonzero when $0\leq x\leq2$ and $1\leq w-x\leq2$. The second condition is equivalent to $w-2\leq x\leq w-1$. Thus the density is one-half times the length of the overlap of $[0,2]$ and $[w-2,w-1]$.

For $1\leq w\leq2$, the overlap length is $w-1$. For $2\leq w\leq3$, it is $1$. For $3\leq w\leq4$, it is $4-w$. Therefore,

f_W(w)= \begin{cases} (w-1)/2, & 1\leq w\leq2,\\ 1/2, & 2\leq w\leq3,\\ (4-w)/2, & 3\leq w\leq4,\\ 0, & \text{otherwise}. \end{cases}

To find $P(W\leq2.5)$, integrate the first two pieces over the relevant intervals:

\begin{aligned} P(W\leq2.5) &=\int_1^2 \frac{w-1}{2}\,dw+\int_2^{2.5}\frac12\,dw\\ &=\left[\frac{(w-1)^2}{4}\right]_1^2+\frac12(2.5-2)\\ &=\frac14+\frac14=0.5. \end{aligned}

A check is that the density integrates to one: the two triangular areas are $1/4$ each, and the central rectangle has area $1/2$, for a total of $1$. Also, $E[W]=E[X]+E[Y]=1+1.5=2.5$; the density is symmetric about $2.5$, consistent with that mean.

Check Distributional Closure and Parameterization

A closure rule is valid only for the specified family and parameterization. For example, independent exponential variables with a common rate have a gamma sum, but exponentials with different rates do not have an exponential sum. An exponential distribution may be parameterized by its rate $\lambda$, giving density $\lambda e^{-\lambda x}$, or by its mean $1/\lambda$. Confusing these quantities changes the distribution.

Worked Example: Different Exponential Rates Do Not Add as Rates

Let $X$ and $Y$ be independent, with $X$ exponential of rate $2$ and $Y$ exponential of rate $1$. Find the density of $W=X+Y$ and calculate $P(W\leq\ln 2)$. Method: use convolution with the stated rate parameterization. Mistake avoided: adding unequal exponential rates and treating the sum as exponential.

The densities are $f_X(x)=2e^{-2x}$ and $f_Y(y)=e^{-y}$ for nonnegative arguments. Since both variables are nonnegative, $W$ has support $[0,\infty)$. For $w\geq0$, convolution gives

\begin{aligned} f_W(w) &=\int_0^w 2e^{-2x}e^{-(w-x)}\,dx\\ &=2e^{-w}\int_0^w e^{-x}\,dx\\ &=2e^{-w}(1-e^{-w})\\ &=2(e^{-w}-e^{-2w}). \end{aligned}

The density is zero for $w<0$. Integrating the density from $0$ to $t$ gives the distribution function for $t\geq0$:

F_W(t)=\int_0^t 2(e^{-w}-e^{-2w})\,dw =1-2e^{-t}+e^{-2t}.

Since $e^{-\ln 2}=1/2$,

P(W\leq\ln 2) =1-2\left(\frac12\right)+\left(\frac12\right)^2 =\frac14.

The incorrect shortcut would be to add the rates and claim $W$ is exponential with rate $3$. That would give $P(W\leq\ln2)=1-e^{-3\ln2}=1-1/8=7/8$, which disagrees with the convolution result. A further check is $E[W]=1/2+1=1.5$, whereas an exponential distribution with rate $3$ would have mean $1/3$. Both checks reject the shortcut.

A Reliable Exam Check

Before accepting a sum distribution, verify the assumptions, the possible values, and the calculation itself. In particular, a density formula must be zero outside the sum's valid support, integrate to one, and agree with basic moment calculations when those moments are available. These checks often catch an answer that has the right-looking family name but the wrong model.

Confirm the joint model.
Use independence only when it is stated or follows from the model. Marginal distributions alone do not establish it.

Find the sum's possible values.
Use the joint model to determine which component values can occur together. Under independence, add the support intervals' endpoints when the stated supports are intervals.

Choose a valid calculation.
Use convolution for independent variables unless a standard closure result applies with matching assumptions and parameters.

Check the result.
Verify support and total probability, then compare means or other moments when available.

A common wrong calculation is to set $f_W(w)=f_X(w)f_Y(w)$. That multiplies two marginal density values at the same argument; it does not account for the different pairs $(x,w-x)$ that can produce the sum $w$. The correction, for independent continuous variables, is $f_W(w)=\int f_X(x)f_Y(w-x)\,dx$, with the integral restricted to valid component values. Another common error is to apply this independent convolution to dependent variables: no such correction is valid without an independence assumption or a specified joint density.

Key takeaway: A sum-distribution calculation is only as reliable as its joint-model assumption. Establish dependence, determine valid support, respect the distribution's parameterization, and use convolution when a closure rule does not apply.

Check Your Understanding

For each problem, identify the relevant model feature before calculating.

Independent discrete variables have $P(X=0)=0.4$, $P(X=1)=0.6$, $P(Y=0)=0.25$, and $P(Y=2)=0.75$. Find $P(X+Y=2)$. (Answer: 0.3000)
Independent variables are uniform on $[0,1]$ and $[0,3]$, respectively. Find the probability that their sum is at most $1$. (Answer: $1/6$, or 0.1667)
Independent exponential variables have rates $4$ and $2$. Give the density of their sum for $w\geq0$. (Answer: $f_W(w)=4(e^{-2w}-e^{-4w})$ for $w\geq0$, and $0$ otherwise)
Independent Bernoulli variables have success probabilities $0.3$ and $0.6$. Find the probability that their sum is exactly $1$. (Answer: 0.5400)
Let $X\sim\operatorname{Uniform}(-1,1)$ and $Y=-X$. Find the support of $X+Y$ and $P(X+Y=0)$. (Answer: support $\{0\}$; probability 1)

What You'll Learn