5
$\begingroup$

Take $w,z\in R^{n}$. I am interested in integrating (as generically if possible) $$\int_{\Omega}(w \cdot z)^{1-\sigma} d z$$ Where the domain of $\Omega$ is $1$ dimension, and includes the convex combinations of $z$. i.e. $\Omega \equiv \{z | \sum z = 1 \}$

Obviously, for $n=2$, this is easy to rearrange: $$\int_{0}^{1} \left(w_1 z_1 + w_2 (1-z_1)\right) d z_1$$ $$ = \frac{w_2^{2-\sigma} - w_1^{2-\sigma}}{(\sigma - 2)(w_1 - w_2)}$$

And the mathematica code is:

Integrate[(w[1] z[1] + w[2] (1 - z[1]))^(1 - σ), {z[1], 0, 1}] // FullSimplify

But it is a little harder to cheat like this as $n$ gets larger, and I am hoping for more general patterns. How would I set up the simple $n=2$ example using the convex combination directly? I tried things like:

 Integrate[(w[1] z[1] + w[2] z[2])^(1 - σ), {z[1], 0, 1}, {z[2], 1 - z[1], 1 - z[1]}] // FullSimplify

Integration over a convex combination of a region: $\int_{\Omega} (w_1 z_1 + w_2 z_2)^{1-\sigma} d (z_1, z_2)$ where $\Omega = \{ z_1 + z_2 = 1\}$

...but I couldn't get them to work.

Improve this question
$\endgroup$
  • $\begingroup$ This will help with generating limits and dummy variables. I'm afraid I don't understand the question enough to help with the integrand. $\endgroup$ – Timothy Wofford Oct 9 '13 at 17:31
  • $\begingroup$ Thanks. Alas, the link you have doesn't seem to have the bounds dropping a dimension, as my example does. $\endgroup$ – jlperla Oct 9 '13 at 17:45
  • $\begingroup$ Well, we can always modify that example. Do you have an equation for the limits? Maybe you could throw a Boole[Total[z]==1] in the integrand. $\endgroup$ – Timothy Wofford Oct 9 '13 at 17:49
  • $\begingroup$ Along the same lines as @TimothyWofford, you might try the appropriate DiracDelta. $\endgroup$ – evanb Apr 15 '14 at 17:40
2
$\begingroup$

Here we enumerate the all the distinct edges in a fully connected graph (an equivalent problem, I believe). Then we parameterize the edge by {t,0,1} assuming that the direction of integration doesn't matter. Then we integrate over each edge and total the integrals.

n = 4;
z[n_] := ToExpression[ToString[z] <> ToString[n]]
edges=Union@Map[Sort, Permutations[Array[z, n], {2}]];
lines = ReplaceAll[Array[z, n], {#[[1]] -> t, #[[2]] -> (1 - t)}] & /@ edges;
(* h[x_]:=x^(1-σ) *)
g[line_]:= h[Array[w, n].line]
Total@Map[Integrate[g[#], {t, 0, 1}] &, lines]

What follows below was an initial answer based on a misunderstanding. I am leaving it because it may be a useful reference.

If I understand correctly, then the limits can be assigned as follows

n = 2;
f[1] := {z[1], 0, 1}
f[n_] := {z[n], 0, 1 - z[n - 1]};
bounds = Array[f, n];
Integrate[(Array[w, n].Array[z, n])^(1 - σ),Sequence @@ bounds]

For those who program better than I do, how can we make this faster? for n=3 I get 4.13607 seconds!

n=3;
bounds = Array[f, n];
Integrate[g[Sequence @@ bounds[[All, 1]]], Sequence @@ bounds]

This brings it down to 2.6 seconds for n=3.

z[n_] := ToExpression[ToString[z] <> ToString[n]]
Improve this answer
$\endgroup$
  • $\begingroup$ Very helpful. But is the first one doing a triangular domain of dimension $n$ rather than of a 1 dimension subspace of the convex combinations?? The Boole[Total[z]==1] mentioned above seems to be the right domain choice $\endgroup$ – jlperla Oct 9 '13 at 18:21
  • $\begingroup$ @jlperla, Aahh, I see. I did a RegionPlot of Boole[Total[z]==1] and misinterpreted the triangle as the volume underneath, rather than the edges. So, we need to parameterize the edge of this triangle, and do a single integration over the parameter? How many edges are there for n=4,5,etc.? $\endgroup$ – Timothy Wofford Oct 9 '13 at 18:26
  • $\begingroup$ @jlperla, I have edited to integrate over the edges. $\endgroup$ – Timothy Wofford Oct 9 '13 at 19:51
  • $\begingroup$ Thank you, I think that does it. $\endgroup$ – jlperla Oct 9 '13 at 20:18
  • $\begingroup$ @TimothyWofford Could you please complete your answer with a definition for line? Regards. $\endgroup$ – chris Apr 10 '15 at 7:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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