0

Let $ x \ge b \ge 0$ and let $\mu_0 \le 1/2$. Then let $j \in {\mathbb R}$. Consider the following integral:

\begin{equation} {\mathcal J}^{(x,b,\mu_0)}_j := \frac{2^{\frac{j}{2}-1}}{\sqrt{\pi}} \cdot \int\limits_0^\infty t^{-\frac{j}{2}-1} \cdot e^{- \frac{\left( \log[\frac{b}{x}] - (\mu_0-\frac{1}{2}) t\right)^2}{2 t}} \cdot H_{j+1} \left[ \frac{\log[\frac{x}{b}]}{\sqrt{2 t} } \right] dt \tag{1} \end{equation} where $H_n[x]$ are Hermite polynomials.

By doing numerical experiments we have found that:

\begin{equation} {\mathcal J}^{(x,b,\mu_0)}_j = \left(1 - 2 \mu_0\right)^j \tag{2} \end{equation}

In[3369]:= LL = {}; M = 2;
Do[
  {x, b} = RandomReal[{0, M}, 2, WorkingPrecision -> 50]; 
  If[b > x, tmp = x; x = b; b = tmp;];
  mu0 = RandomReal[{-M, 1/2}, WorkingPrecision -> 50];
  j = RandomReal[{-M, M}, WorkingPrecision -> 50];
  ll = {1/Sqrt[Pi] NIntegrate[
      2^(j/2 - 1) (t)^(-j/2 - 1)
        Exp[-((Log[b/x] - (mu0 - 1/2) t)^2/( 2 t))] HermiteH[j + 1, 
        Log[x/b]/(Sqrt[2] t^(1/2))], {t, 0, Infinity}, 
      WorkingPrecision -> 20], (1 - 2 mu0)^j};
  LL = Join[LL, {ll}];
  , {which, 1, 100}];
Mean[Abs[#[[1]]/#[[2]] - 1] & /@ LL]

Out[3371]= 2.425121*10^-13

Being intrigued by the simplicity of this result we tried to prove it.

Below, we prove it for $ j \in {\mathbb Z}$ and $j \ge -1$. Here we go:

\begin{eqnarray} &&{\mathcal J}^{(x,b,\mu_0)}_j = \\ && \frac{1}{\sqrt{\pi}} \exp \left[ (-\frac{1}{2} + \mu_0) \log[\frac{b}{x}]\right] \cdot \sum\limits_{m_2=0}^{\lfloor \frac{j+1}{2} \rfloor} \frac{(-1)^{-m_2} (j+1)! 2^{j-m_2-\frac{1}{2}} }{m_2! \left(j-2 m_2+1\right)!} [\log \left(\frac{x}{b}\right)]^{j-2 m_2+1} % \int\limits_0^\infty t^{-j+m_2-\frac{3}{2}}e^{-\frac{\log ^2\left(\frac{b}{x}\right)}{2 t}-\frac{1}{8} \left(1-2 \mu _0\right){}^2 t} dt = \\ && \frac{1}{\sqrt{\pi}} \exp \left[ (-\frac{1}{2} + \mu_0) \log[\frac{b}{x}]\right] \cdot \sum\limits_{m_2=0}^{\lfloor \frac{j+1}{2} \rfloor} \frac{(-1)^{-m_2} (j+1)! }{m_2! \left(j-2 m_2+1\right)!} % \left(\frac{\log ^{\frac{1}{2}-m_2}\left(\frac{x}{b}\right)}{(1-2 \mu _0)^{j-m_2+\frac{1}{2}}}\right) K_{-j+m_2-\frac{1}{2}}\left(\log \left(\frac{x}{b}\right) \left(\frac{1}{2}-\mu _0\right)\right) = \\ && \left( 1 - 2 \mu_0\right)^j \cdot \left[ % 1_{j=-1} + 1_{j \ge 0} \sum\limits_{L=0}^j \left( \log[\frac{x}{b}] (1-2 \mu_0)\right)^{-L} \cdot \underbrace{ \frac{(j+1)!}{(-L+j)!} \cdot \sum\limits_{m_2=0}^{\lfloor \frac{j+1}{2} \rfloor \wedge L} \frac{(-1)^{-m_2} \left(j+L-2 m_2\right)!}{m_2! \left(j-2 m_2+1\right)! \left(L-m_2\right)!} }_{\delta_{L,0}} % \right] = \\ && \left( 1 - 2 \mu_0\right)^j \end{eqnarray} In the second line we used Rodrigues' formula and expanded the Hermite polynomials and we swapped the integration with the sum over the terms. In the third line we used (Gradshteyn & Ryzhik 3.471.9) at $\alpha = 1/2 \log[b/x]^2$, $\beta = 1/8 (1-2 \mu_0)^2$ and $\nu = -1/2-j+m_2$. In the forth line we used this result for the modified Bessel functions of a half-integer argument.

(*Check normalization. (*j\[GreaterEqual]-1;*)*)
{x, b} = RandomReal[{0, 2}, 2, WorkingPrecision -> 50]; If[b > x, 
 tmp = x; x = b; b = tmp;];
mu0 = RandomReal[{-2, 0}, WorkingPrecision -> 50];
n = RandomInteger[{1, 10}]; th = 9/10;
j = RandomInteger[{-1, n}];(*j\[GreaterEqual]-1;*)

1/Sqrt[Pi] NIntegrate[ 2^(j/2 - 1) (t)^(-j/2 - 1) Exp[-((Log[b/x] - (mu0 - 1/2) t)^2/( 2 t))] HermiteH[j + 1, Log[x/b]/(Sqrt[2] t^(1/2))], {t, 0, Infinity}]

1/Sqrt[Pi] Exp[+(-(1/2) + mu0) Log[b/ x]] Sum[(j + 1)!/((j + 1 - 2 m2)! m2!) 2^(j - 1/2 - m2) (-1)^(-m2) ( Log[x/b])^(j + 1 - 2 m2) NIntegrate[ t^(-(3/2) - j + m2) Exp[-(1/8) (1 - 2 mu0)^2 t - Log[b/x]^2/(2 t)], {t, 0, Infinity}], {m2, 0, Floor[(j + 1)/2]}]

(Use https://math.stackexchange.com/questions/128687/bessel-function-
integral-and-mellin-transform?rq=1 (Gradshteyn&Ryzhik 3.471.9,) at
alpha =Log[b/x]^2/2, beta = 1/8 (1-2 mu0)^2, nu = -1/2-j+m2
) 1/Sqrt[Pi] Exp[+(-(1/2) + mu0) Log[b/ x]] Sum[(j + 1)!/((j + 1 - 2 m2)! m2!) (-1)^(-m2) ( Log[x/b])^(1 - m2 - 1/2) (1/(1 - 2 mu0))^((-1/2 - j + m2)) BesselK[-1/2 - j + m2, Log[x/b] (1/2 - mu0)], {m2, 0, Floor[(j + 1)/2]}] (Use https://functions.wolfram.com/Bessel-TypeFunctions/BesselK/03/
01/02/02/0001/
) (1 - 2 mu0)^j If[j == -1, 1, Sum[(j + 1)!/(-L + j)! Sum[((-1)^(-m2) (+L + j - 2 m2)! )/((j + 1 - 2 m2)! m2! (L - m2)! ), {m2, 0, Min[Floor[(j + 1)/2], L]}] (Log[x/b] (1 - 2 mu0))^(-L) , {L, 0, j}]]

((1 - 2 mu0)^j)

enter image description here


Having said all this my question is how do we prove $(2)$ for other values of $j$ in particular for negative integers. Clearly Hermite polynomials are well defined for the order being a negative integer as explained here for example.

Update:

By using the same considerations as above one can show that proving the identity in question is equivalent to proving that: $\sum\limits_{n=0}^\infty \frac{(-1)^{n+1} 2^{\frac{j-n}{2}} A^{\frac{n-j}{2}} (j-n+2)_n \Gamma \left(\frac{j-n}{2}+1\right) \left(I_{\frac{1}{2} (-j-n)}(A)-I_{\frac{j+n}{2}}(A)\right)}{n!} = \exp(-A)$. Here $A \ge 0 $ and $j \in {\mathbb R}$. and $I_{\cdot}(\cdot)$ is the modified Bessel function of the first kind.

Przemo
  • 11,331

1 Answers1

1

This is by no means a complete answer to this question but I believe it is a step in the right direction so I am posting it. If we start from the conjectured identity given in the Update above and use the series representation of the modified Bessel function of the first kind then we need to prove the following:

\begin{equation} (-1) \sum\limits_{n=0}^\infty \frac{(-j-1)^{(n)} \Gamma[1+\frac{j}{2} - \frac{n}{2}]}{n!} \cdot \sum\limits_{m=0}^\infty \frac{(A/2)^{2 m}}{m!} \left[ \frac{(A/2)^{-j}}{(m - \frac{j+n}{2})!} - \frac{(A/2)^{+n}}{(m + \frac{j+n}{2})!}\right] = \exp(-A) \tag{A} \end{equation}

Now by swapping the order of sums in the above and doing the sum over $n\in {\mathbb N}$ the identity $(A)$ is equivalent to the following two identities:

\begin{eqnarray} &&\sum\limits_{m=0}^\infty \frac{(A/2)^{2 m}}{m!} \cdot \\ &&\left[ \frac{\Gamma \left(\frac{j+2}{2}\right) \, _1F_2\left(-\frac{j}{2}-\frac{1}{2};\frac{1}{2},\frac{j}{2}+m+1;-\frac{A^2}{4}\right)}{\left(\frac{j}{2}+m\right)!}-\frac{A \Gamma \left(\frac{j+3}{2}\right) \, _1F_2\left(-\frac{j}{2};\frac{3}{2},\frac{j}{2}+m+\frac{3}{2};-\frac{A^2}{4}\right)}{\left(\frac{1}{2} (j+2 m+1)\right)!}\right] = \\ && \exp(-A) \tag{B1} \end{eqnarray}

and

\begin{eqnarray} &&\sum\limits_{m=0}^\infty \frac{(A/2)^{2 m}}{m!} \cdot \\ &&2^j A^{-j} \left(\frac{\Gamma \left(\frac{j+2}{2}\right) \, _2F_1\left(-\frac{j}{2}-\frac{1}{2},\frac{j}{2}-m;\frac{1}{2};1\right)}{\left(m-\frac{j}{2}\right)!}-\frac{2 \Gamma \left(\frac{j+3}{2}\right) \, _2F_1\left(-\frac{j}{2},\frac{1}{2} (j-2 m+1);\frac{3}{2};1\right)}{\left(-\frac{j}{2}+m-\frac{1}{2}\right)!}\right) = \\ && 0 \tag{B2} \end{eqnarray}

But now if we use Mathematica and we expand the left hand sides of the conjectured identities $(B1)$ and $(B2)$ in a series in the variable $A$ then we get the following:

In[3892]:= 
M = 5; j =.; m =.; A =.;
Collect[Normal[
  Series[Sum[(A/2)^(2 m)/
      m! ((Gamma[(2 + j)/
         2] HypergeometricPFQ[{-(1/2) - j/2}, {1/2, 1 + j/2 + m}, -(
          A^2/4)])/(j/2 + m)! - (
       A  Gamma[(3 + j)/
         2] HypergeometricPFQ[{-(j/2)}, {3/2, 3/2 + j/2 + m}, -(A^2/
          4)])/ (1/2 (1 + j + 2 m))!), {m, 0, M}], {A, 0, 
    2 M}]], A, FullSimplify]
Collect[Normal[
  Series[Sum[(A/2)^(2 m)/
      m! (2^j A^-j ((
         Gamma[(2 + j)/2] Hypergeometric2F1[-(1/2) - j/2, j/2 - m, 1/
           2, 1])/(-(j/2) + m)! - (
         2 Gamma[(3 + j)/2] Hypergeometric2F1[-(j/2), 
           1/2 (1 + j - 2 m), 3/2, 1])/(-(1/2) - j/2 + m)!)), {m, 0, 
     M}], {A, 0, 2 M}]], A, FullSimplify]

Out[3893]= 1 - A + A^2/2 - A^3/6 + A^4/24 - A^5/120 + A^6/720 -
A^7/5040 + A^8/40320 - A^9/362880 + A^10/3628800

Out[3894]= 0

And this result is exactly what we would be expecting. The first ouput is just a truncated Taylor expansion of $\exp(-A)$!.

So what remains is to prove that $(B1)$ and $(B2)$ are identities.

Update:

Here we prove that $(B1)$ is an identity. We will show that coefficients at particular powers in both sides are equal. Here we go.

Even powers:

\begin{eqnarray} &&@[A^{2 m}](lhs)=\\ && \frac{1}{2^{2 m}} \sum\limits_{l=0}^m \frac{1}{(m-l)!} \frac{(-1)^l}{l!} \cdot \frac{(-j/2-1/2)^{(l)}}{(1/2)^{(l)} (j/2+m-l+1)^{(l)} } \cdot \frac{\Gamma(j/2+1)}{(j/2+m-l)!}= \\ && \frac{1}{2^{2 m}} \sum\limits_{l=0}^m \frac{1}{(m-l)!} \frac{(-1)^l}{l!} \cdot \frac{(-j/2-1/2)^{(l)}}{(1/2)^{(l)} \underline{ (j/2+m-l+1)^{(l)} (j/2+1)^{(m-l)} }}= \\ && \frac{1}{2^{2 m}} \sum\limits_{l=0}^m \frac{1}{(m-l)!} \frac{(-1)^l}{l!} \cdot \frac{(-j/2-1/2)^{(l)}}{(1/2)^{(l)} \underline{ (j/2+1)^{(m)} }}= \\ && \frac{1}{2^{2 m}} \frac{1}{ (j/2+1)^{(m)} } \sum\limits_{l=0}^m \frac{1}{(m-l)!} \frac{(-1)^l}{l!} \cdot \frac{(-j/2-1/2)^{(l)}}{(1/2)^{(l)} }= \\ && \frac{1}{2^{2 m}} \frac{1}{ (j/2+1)^{(m)} } \cdot \frac{2^{2 m} \left(\frac{1}{2} (j+2 m)\right)!}{\frac{j}{2}! (2 m)!} = \\ && \frac{1}{2^{2 m}} \cdot \frac{2^{2 m} }{ (2 m)!} = \\ && \frac{1}{ (2 m)!} = \\ &&@[A^{2 m}](rhs) \end{eqnarray}

Odd powers:

\begin{eqnarray} &&@[A^{2 m+1}](lhs)=\\ && -\frac{1}{2^{2 m}} \sum\limits_{l=0}^m \frac{1}{(m-l)!} \frac{(-1)^l}{l!} \cdot \frac{(-j/2)^{(l)}}{(3/2)^{(l)} (j/2+m-l+3/2)^{(l)}} \cdot \frac{\Gamma((j+3)/2)}{\Gamma((j+3)/2 + m-l)} = \\ && -\frac{1}{2^{2 m}} \sum\limits_{l=0}^m \frac{1}{(m-l)!} \frac{(-1)^l}{l!} \cdot \frac{(-j/2)^{(l)}}{(3/2)^{(l)} \underline{ (j/2+m-l+3/2)^{(l)} ((j+3)/2)^{(m-l)} }} = \\ && -\frac{1}{2^{2 m}} \sum\limits_{l=0}^m \frac{1}{(m-l)!} \frac{(-1)^l}{l!} \cdot \frac{(-j/2)^{(l)}}{(3/2)^{(l)} \underline{ ((j+3)/2)^{(m)} }} = \\ && -\frac{1}{2^{2 m}} \frac{1}{ ((j+3)/2)^{(m)} } \sum\limits_{l=0}^m \frac{1}{(m-l)!} \frac{(-1)^l}{l!} \cdot \frac{(-j/2)^{(l)}}{(3/2)^{(l)} } = \\ && -\frac{1}{2^{2 m}} \frac{1}{ ((j+3)/2)^{(m)} } \frac{2^{2 m} \left(\frac{1}{2} (j+2 m+1)\right)!}{\frac{j+1}{2}! (2 m+1)!} = \\ && -\frac{1}{2^{2 m}} \frac{2^{2 m} }{ (2 m+1)!} = \\ &&- \frac{1}{ (2 m+1)!} = \\ && @[A^{2m+1}](rhs) \end{eqnarray}

Finally we prove that $(B2)$ is an identity as well. Again, we look at the coefficients of the left hand side at powers of $A$. We have:

\begin{eqnarray} &&@[A^{2 m}](lhs)=\\ &&\frac{1}{2^{2 m}} \frac{1}{m!} 2^{j} A^{-j} \cdot \left[ \right. \\ && \left. \frac{ \Gamma(\frac{j+2}{2}) \hspace -5ex\diagup }{(m- \frac{j}{2})!} \cdot \underline{ \frac{\Gamma(\frac{1}{2}) \Gamma(\frac{1}{2} + \frac{1}{2} +m)}{\Gamma(1+\frac{j}{2}) \hspace -5ex\diagup \hspace 5ex \Gamma(\frac{1}{2} - \frac{j}{2} +m)} } \right. \\ && \left. - 2 \cdot \frac{\Gamma(\frac{j+3}{2}) \hspace -5ex\diagup}{(-\frac{j}{2} + m - \frac{1}{2})!} \cdot \underline{ \frac{\Gamma(\frac{3}{2}) \Gamma(\frac{3}{2} + m - \frac{1}{2})}{\Gamma(\frac{3}{2} + \frac{j}{2}) \hspace -5ex\diagup \hspace 5ex \Gamma(\frac{3}{2} - \frac{j}{2} + m -\frac{1}{2})} } \right. \\ && \left. \right] = \\ %%%%% &&\frac{1}{2^{2 m}} \frac{1}{m!} 2^{j} A^{-j} \cdot \left[ \right. \\ && \left. \frac{1}{(m- \frac{j}{2})!} \cdot \underline{ \frac{\Gamma(\frac{1}{2}) \Gamma(\frac{1}{2} + \frac{1}{2} +m)}{ \Gamma(\frac{1}{2} - \frac{j}{2} +m)} } \right. \\ && \left. - 2 \cdot \frac{1}{(-\frac{j}{2} + m - \frac{1}{2})!} \cdot \underline{ \frac{\Gamma(\frac{3}{2}) \Gamma(\frac{3}{2} + m - \frac{1}{2})}{\Gamma(\frac{3}{2} - \frac{j}{2} + m -\frac{1}{2})} } \right. \\ && \left. \right] = \\ %%%%%% &&\frac{1}{2^{2 m}} \frac{1}{m!} 2^{j} A^{-j} \cdot \left[ \right. \\ && \left. \frac{1}{(m- \frac{j}{2})!} \cdot \underline{ \frac{\Gamma(\frac{1}{2}) \Gamma(1 +m)}{ \Gamma(\frac{1}{2} - \frac{j}{2} +m)} } \right. \\ && \left. - 2 \cdot \frac{1}{\Gamma(-\frac{j}{2} + m + \frac{1}{2})} \cdot \underline{ \frac{\Gamma(\frac{3}{2}) \Gamma(1 + m )}{( - \frac{j}{2} + m )!} } \right. \\ && \left. \right] = \\ %%%%%%% && 0 \end{eqnarray}

In the first line we used the Gauss' summation theorem and then we just simplified.

Przemo
  • 11,331