2
$\begingroup$

How would I tell Mathematica to evaluate something like

\begin{align*} \exp\left[\int_a^b dx\right]f(x)\equiv f(x)+\int_a^b f(x)dx+\frac{1}{2!}\int_a^b\int_0^{x}f(x')dx'dx+\frac{1}{3!}\int_a^b\int_0^{x}\int_0^{x'}f(x'')dx''dx'dx+\cdots \end{align*}

If $f(x)$ was a polynomial, then this would be straightforward, for each integral would be known and the function of Sum could then be used. However, in general, $f(x)$ will not be that simple. I was thinking that a similar method to the one described in Application of # for the $n^{th}$ derivative could be used, but I am not sure how to apply it for integrals.

If Mathematica is unable to do the exponential of the integral operator, is it possible to compute the series expansion up to a specified term?

$\endgroup$
  • 1
    $\begingroup$ Because f is a function of the innermost variable only, it should be possible systematically to interchange the order of integration, one level at a time, to collapse all integrals in each term of the series into a single integral, multiplied by a polynomial in a and b. Once the general term has been obtained (inductively), the series probably can be summed. Unfortunately, Integrate apparently cannot handle this process automatically. $\endgroup$ – bbgodfrey Aug 31 '16 at 22:29
  • $\begingroup$ Repeated integrals of the form you are using can be written as a single integral. I think that symbolic manipulation can greatly simplify this. $\endgroup$ – mikado Sep 1 '16 at 5:01
  • $\begingroup$ mathworld.wolfram.com/RepeatedIntegral.html gives a useful result $\endgroup$ – mikado Sep 1 '16 at 5:29
  • $\begingroup$ I am not sure to have correctly understood your need, but is this expOp = Exp[Integrate[#, {x, a, b}]] & not what you are looking for? This expOp[x] gives E^(-(a^2/2) + b^2/2) and this expOp[Cos[x]^2] yields E^(1/2 (-a + b - Cos[a] Sin[a] + Cos[b] Sin[b])) ?? $\endgroup$ – Alexei Boulbitch Sep 1 '16 at 6:56
  • 1
    $\begingroup$ Are you looking for a numerical solution? Then one could adapt my answer here, because this is essentially the Picard iteration except for the factorials. $\endgroup$ – Jens Sep 3 '16 at 16:11
1
$\begingroup$

Let us start by defining 3 alternative ways of performing the multiple integral

int1[f_, x_, 0] = f[x];
int1[f_, x_, n_] := Integrate[int1[f, u, n - 1], {u, 0, x}]
int2[f_, x_, n_] := Derivative[-n][f][x]
int3[f_, x_, 0] = f[x];
int3[f_, x_, n_] := Integrate[f[u] (x - u)^(n - 1)/(n - 1)!, {u, 0, x}]

The first corresponds to a direct iteration of the integral, the second use of the antiderivative properties of Derivative and the third exploits the transform given at http://mathworld.wolfram.com/RepeatedIntegral.html To test the behaviour of these apply them to Sin. We see that the first and third agree, whereas the second does not handle the end conditions correctly.

Table[Through[{int1, int2, int3}[Sin, x, i]], {i, 0, 3}] // 
  Expand // InputForm

(* {{Sin[x], Sin[x], Sin[x]}, {1 - Cos[x], -Cos[x], 
  1 - Cos[x]}, {x - Sin[x], -Sin[x], 
  x - Sin[x]}, {-1 + x^2/2 + Cos[x], Cos[x], -1 + x^2/2 + Cos[x]}} *)

The integral of interest takes the form

Sum[a[i] int[f, x, i], {i, 0, ∞}];

using the repeated integral expression we have

% /. u_int :> int3 @@ u // InputForm
(* Sum[a[i]*Integrate[((-u + x)^(-1 + i)*f[u])/(-1 + i)!, {u, 0, x}], {i, 0, Infinity}] *)

which can be rewritten

Integrate[
  f[u]*Sum[a[i]*((-u + x)^(-1 + i)/(-1 + i)!), {i, 0, Infinity}], {u, 
   0, x}];

substituting for the required expression we get

% /. a[i_] -> 1/i! // InputForm
(* Integrate[(BesselI[1, 2*Sqrt[-u + x]]*f[u])/Sqrt[-u + x], {u, 0, x}] *)
$\endgroup$

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.