Problem 3.6. Recall Eq (3.4):

where and .

Show that is Gaussian:

Per Bayes’ theorem , and our goal is to massage the right-hand side into the standard Gaussian form

(0) First let us remind ourselves how the change of variables work.

When we have a continuous measure then we can write where is the density function with respect to the Lebesgue measure. We can write . Now let us look at some transformation and we care about .

Let be measurable and function and . We say that is measurable or -measurable if every Borel set has it’s original in .

Now let us look at measure spaces and . Let us define a generalisation for .

Let us have a function such that for every . Now define

If then . If , then for every we have

Now specialise this machinery to densities — which is all the “Jacobian” ever was. Let be a random variable on with law on , and suppose has a density with respect to Lebesgue measure , i.e. . Take a diffeomorphism and set . Pushforwards compose, so the law of is

For any , the image-measure identity (*) with gives

So far there is no Jacobian: the abstract change of variables (*) is exact for any measurable . The factor enters only when we re-express this integral against itself, via the substitution . That substitution is not imported from calculus as a black box — it is justified, within this same framework, by the following lemma.

Lemma (density of the pushforward of Lebesgue measure). For a diffeomorphism , the pushforward is absolutely continuous with respect to , with

Proof. Take increasing (the decreasing case is identical up to a sign). For an interval we have , so by the fundamental theorem of calculus (valid since ),

For decreasing the endpoints swap and , which is exactly what produces the absolute value. Both sides are measures in agreeing on the -system of intervals, hence agree on all of by uniqueness of measures.

Now combine the two. Apply (*) with , so that :

Trying to derive the first equation in prev eq.

the last equality by the lemma. This is the substitution rule in differential shorthand; the only analytic input was the fundamental theorem of calculus applied to on intervals. Taking turns our expression for the law of into

Since this holds for every , uniqueness of the Radon–Nikodym derivative identifies the density of :

The factor is precisely the Radon–Nikodym derivative — the density with which distorts Lebesgue measure. That is the whole content of the word “Jacobian”: not a separate rule for densities, but the price of measuring against a that does not preserve.

For the affine map (with ) we have and , so

With , this is the used next; for it returns .

(1) Recall the convolution operation. For standard normal rvs X,Y we have

X+Y should have and

Let it be standard normal so we simplify to

Complete the square in to separate the part from the integration variable:

so

The remaining integral is a plain Gaussian, , independent of . Hence

which is exactly , confirming and .

(2) Apply the affine rule from (0) with , :

which is : the from (0) turns the prefactor into , exactly the normalization for standard deviation .

Now assemble the three pieces. Since with , conditioning on shifts the mean by , so . The prior is , and the evidence is a plain number (here the is the factor from the affine rule in (0)):

Multiply likelihood and prior and complete the square in :

and since (using and ),

Now divide by — the algebra closes on its own, no appeal to “it must be a normalizer.” The factors cancel exactly, and the prefactor collapses:

so

This is exactly the density of , correctly normalized by direct computation — the dividing supplies precisely the that a Gaussian requires.