# Expectation operator - linearity

1. Dec 8, 2012

### Pietair

1. The problem statement, all variables and given/known data
Show that the expectation operator E() is a linear operator, or, implying:
$$E(a\bar{x}+b\bar{y})=aE(\bar{x})+bE(\bar{y})$$

2. Relevant equations
$$E(\bar{x})=\int_{-\infty}^{+\infty}xf_{\bar{x}}(x)dx$$

With $$f_{\bar{x}}$$ the probability density function of random variable x.

3. The attempt at a solution
$$aE(\bar{x})=a\int_{-\infty}^{+\infty}xf_{\bar{x}}(x)dx$$ and:
$$bE(\bar{y})=b\int_{-\infty}^{+\infty}yf_{\bar{y}}(y)dy$$

Introducing a new random variable:
$$\bar{v}=a\bar{x}+b\bar{y}$$

Then:

$$E(\bar{v})=E(a\bar{x}+b\bar{y})=\int_{-\infty}^{+\infty}vf_{\bar{v}}(v)dv=\int_{-\infty}^{+\infty}(ax+by)f_{\bar{v}}(v)dv$$

And accordingly:
$$E(a\bar{x}+b\bar{y})=\int_{-\infty}^{+\infty}(ax+by)f_{\bar{v}}(v)dv=a\int_{-\infty}^{+\infty}xf_{\bar{v}}(v)dv+b\int_{-\infty}^{+\infty}yf_{\bar{v}}(v)dv$$

So what remains to proof is that:

$$a\int_{-\infty}^{+\infty}xf_{\bar{v}}(v)dv+b\int_{-\infty}^{+\infty}yf_{\bar{v}}(v)dv=a\int_{-\infty}^{+\infty}xf_{\bar{x}}(x)dx+b\int_{-\infty}^{+\infty}yf_{\bar{y}}(y)dy$$

And now I am stuck... I don't know how I can relate the p.d.f. of random variable v to the p.d.f.'s of random variables x and y.

2. Dec 8, 2012

### micromass

Staff Emeritus
Haven't you seen some kind of results that gives the pdf of X+Y in terms of the pdf of X and Y?? Hint: it has to do with convolution.

3. Dec 8, 2012

### Pietair

I know that the probability distribution of the sum of two or more random variables is the convolution of their individual pdf's, but as far as I know this is only valid for independent random variables. While $E(aX+bY)=aE(X)+bE(Y)$ is true in general, right?.

4. Dec 8, 2012

Yes to both.

5. Dec 8, 2012

### Ray Vickson

Depending on the level of rigor required, you may have to qualify things a bit. For example, if Y = -X, then E(X+Y) = EX + EY is false if EX = EY = ∞, because we would be trying to equate 0 to ∞ - ∞, which is not allowed.

So, the easiest way is to assume that both EX and EY are finite. Now you really have a 2-stage task:
(1) Prove the "theorem of the unconscious statistician", which says that if Z = g(X,Y), then
$$EZ \equiv \int z \: dF_Z(z)$$
can be written as
$$\int g(x,y) \: d^2F_{XY}(x,y),$$
which becomes
$$\sum_{x,y} g(x,y)\: P_{XY}(x,y)$$
in the discrete case where there are no densities, and becomes
$$\int g(x,y) f_{XY}(x,y) \, dx \, dy$$
in the continuous case where there are densities. Of course, the result is also true in a mixed continuous-discrete case where there are densities and point-mass probabilities, but then we need to write Stieltjes integrals, etc. Then, you need to do the much easier task of proving that
$$\int (ax + by) f_{XY}(x,y) \, dx \, dy = a \int x f_X(x) \, dx + b \int y f_Y(y) \, dy = a EX + b EY .$$
The last form holds in general, whether the random variables are continuous, discrete or mixed. As you say, it holds even when X and Y are dependent.

Last edited: Dec 8, 2012
6. Dec 9, 2012

### micromass

Staff Emeritus
Hmm, so there really is no elementary proof?? (I guess it depends on what you call elementary though). This kind of makes me happy that I know measure theory, it makes the proof of this result so much simpler.