4
$\begingroup$

What's the best way to solve the following with $x\sim \text{Normal}(0,1)$?

$$E_x[\log(1-2 a x^2 + a^2 x^4)]=0$$

Idea 1: FindRoot

a0 = 2.42;
term[a_] := 
  TransformedDistribution[Log[1 - 2 a x^2 + a^2 x^4], 
   x \[Distributed] NormalDistribution[]];
FindRoot[NExpectation[t, t \[Distributed] term[a]], {a, a0}]
(* NIntegrate failed to converge to prescribed accuracy *)

Idea 2: increase MaxRecursion, AccuracyGoal

a0 = 2.42;
method = {"NIntegrate", {MinRecursion -> 12, MaxRecursion -> 30, 
    WorkingPrecision -> 30, AccuracyGoal -> 12}};
term[a_] := 
  TransformedDistribution[Log[1 - 2 a x^2 + a^2 x^4], 
   x \[Distributed] NormalDistribution[]];
FindRoot[
 NExpectation[t, t \[Distributed] term[a], Method -> method], {a, a0}]
(* {a -> 2.42125 - 2.16855*10^-9 I} *)

Idea 3: look at density to find tricks

a0 = 2.42;
func = With[{a = a0}, 
   PDF[TransformedDistribution[Log[1 - 2 a x^2 + a^2 x^4], 
     x \[Distributed] NormalDistribution[]]]];
Plot[func[x], {x, -3, 3}]

enter image description here

Now it seems an integral splitting technique would work, but I can't get PDF for symbolic $a$, any ideas?

Background: this gives largest $a$ such that $w=w-a w x^2$ converges almost surely for any starting $w$

$\endgroup$

1 Answer 1

6
$\begingroup$

Part 1. Using $1-2ax^2+a^2x^4 = |1-ax^2|^2$ and the law of the unconscious statistician:

pdfN01=1/Sqrt[2*Pi]*Exp[-x^2/2];
expectationAux:=Integrate[2*Log[Abs[1-a*x^2]]*pdfN01,{x,-Infinity,Infinity}]//FullSimplify;
expectation=Piecewise[{
    {Assuming[a>0,expectationAux],a>0},
    {Assuming[a<0,expectationAux],a<0}},0];
(*
Piecewise[{
  {2*(-EulerGamma + HypergeometricPFQ[{1, 1}, {3/2, 2}, -1/2*1/a]/a + Log[a/2]), a > 0}, 
  {2*(-EulerGamma + Pi*Erfi[1/(Sqrt[2]*Sqrt[-a])] + HypergeometricPFQ[{1, 1}, {3/2, 2}, -1/2*1/a]/a + Log[-1/2*a]), a < 0}},
0]
*)

Plot of the result:

Plot[expectation,{a,-3,8},WorkingPrecision->100,AxesLabel->{"a","Expectation"}]

enter image description here

Part 2. Determine the $a\approx 2.5$ where the expectation vanishes:

FindRoot[expectation,{a,25/10},WorkingPrecision->100]
(* a->2.421249521036836042992780131428225933128842834067867847377838424345397835692066559296652232682365322 *)
$\endgroup$
1
  • $\begingroup$ Neat trick with Piecewise, learning something new every day! $\endgroup$ Commented Aug 21, 2022 at 20:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.