Skip to main content
added 461 characters in body
Source Link
Fedor Petrov
  • 108.9k
  • 9
  • 264
  • 459

Ah, we may simply integrate by parts!

Denote $h(t)=(2 - e^{-t}( 2 + 2t+t^2))(1-e^{-t})^{n-1}$. Integrating by parts $(n+1)$ times we get $$\int_0^\infty h(t)t^{-n-2}dt=\frac1{(n+1)!}\int_0^\infty h^{(n+1)}(t)t^{-1}dt,$$ all off-integral terms vanish since $h(0)=h'(0)=\dots=h^{(n+1)}(0)=0$ and at infinity we have $h(t)=O(1)$, $h^{(i)}(t)=o(1)$ for $i>0$. We have $h^{(n+1)}(t)=\sum_{k=1}^n a_ke^{-kt}+u(t)$, where $u(t)=t\times \text{polynomial}(e^{-t},t)$. Note that $\sum a_k=h^{(n+1)}(0)=0$, thus $$\int_0^\infty \sum a_ke^{-kt}t^{-1}dt=\int_0^\infty \sum a_k(e^{-kt}-e^{-t})t^{-1}dt=-\sum a_k \log k$$ by Frullani integrals. It is easy to see that $$a_k=(-1)^{k+n+1}\binom{n-1}{k-1}k^{n-1}(n^2+n-2k).$$ It remains to evaluate $\int_0^\infty u(t) t^{-1}dt$. We have $$u(t)=-(2t+t^2)(g(t))^{(n+1)}-2(n+1)t(g(t))^{(n)},\,g(t)=e^{-t}(1-e^{-t})^{n-1}.$$ Therefore $$\int_0^\infty u(t) t^{-1}dt=-2\int g^{(n+1)}(t)dt-2(n+1)\int g^{(n)}(t)dt-\int_0^{\infty}tg^{(n+1)}(t)dt.$$ We get $\int (g^{(n)}+tg^{(n+1)})=tg^{(n)}$, and the definite integral against $(0,\infty)$ equals 0. It remains $\int_0^\infty -2g^{(n+1)}-(2n+1)g^{(n)}=2g^{(n)}(0)+(2n+1)g^{(n-1)}(0)$. We have $$g(t)=(1-t+\dots)t^{n-1}(1-t/2+\dots)^{n-1}=t^{n-1}-\frac{n+1}2t^n+\dots,$$ $g^{(n-1)}(0)=(n-1)!$, $g^{n}(0)=-\frac{(n+1)!}2$, $2g^{(n)}(0)+(2n+1)g^{(n-1)}(0)=-(n^2-n-1)(n-1)!$, confirming the guess of Sylvain JULIEN.

PREVIOUS NON COMPLETE VERSION

Still not a complete answer, but a method to be completed or improved.

Denote $f_n(t)=\frac{2 - e^{-t}\left ( 2 + 2t+t^2 \right )}{t^3}\left ( \frac{1-e^{-t}}{t} \right )^{n-1}$. Assume that we have found the coefficients $c_1,c_2,\dots,c_n$ and polynomials $g_1,\dots,g_n$ such that $$f_n(t)=2t^{-n-2}+\sum_{k=1}^n(g_k(t^{-1})e^{-kt})'+\sum_{k=1}^n c_k \frac{e^{-kt}}t.\,\,(*)$$ Then $\sum c_k=0$ (else $f_n$ would have a non-zero residue at 0, which is absurd). We have $\int_0^\infty \sum_{k=1}^n c_k \frac{e^{-kt}}t dt=-\sum c_k\log k$ by Frullani integrals. The integral of the other part of our sum $(*)$ is minus the limit at zero of the function $-\frac2{n+1}t^{-n-1}+\sum g_k(t^{-1})e^{-kt}$. The limit must exist, since the initial integral converges.

Now I cheat a bit. Note that if we write $f_n(t)=\sum_{k=0}^n q_k(1/t) e^{-kt}$ for polynomials $q_k$, then $c_k$ equals to the residue of the $q_k(1/t) e^{-kt}$, which is pretty computable. If I am not mistaken, $$q_k(t^{-1})=(-1)^kt^{-n-2}\left(2\binom{n-1}k+\binom{n-1}{k-1}(2+2t+t^2)\right)=\\=(-1)^kt^{-n-2}\left(2\binom{n}k+\binom{n-1}{k-1}(2t+t^2)\right).$$ Thus $$c_k=2(-1)^{k+n+1}\binom{n}k\frac{k^{n+1}}{(n+1)!}+2(-1)^{k+n}\binom{n-1}{k-1}\frac{k^{n}}{n!}+(-1)^{k+n+1}\binom{n-1}{k-1}\frac{k^{n-1}}{(n-1)!}=\\ =(-1)^{k+n+1}\frac{k^{n-1}(n^2+n-2k)}{n(n+1)(n-k)!(k-1)!}.$$ This matches a coefficient of $\log 3$ for $n=4$ from Shahrooz Janbaz's answer, you may check others for be sure. It remains to prove Sylvain JULIEN's guess for the rational part.

Still not a complete answer, but a method to be completed or improved.

Denote $f_n(t)=\frac{2 - e^{-t}\left ( 2 + 2t+t^2 \right )}{t^3}\left ( \frac{1-e^{-t}}{t} \right )^{n-1}$. Assume that we have found the coefficients $c_1,c_2,\dots,c_n$ and polynomials $g_1,\dots,g_n$ such that $$f_n(t)=2t^{-n-2}+\sum_{k=1}^n(g_k(t^{-1})e^{-kt})'+\sum_{k=1}^n c_k \frac{e^{-kt}}t.\,\,(*)$$ Then $\sum c_k=0$ (else $f_n$ would have a non-zero residue at 0, which is absurd). We have $\int_0^\infty \sum_{k=1}^n c_k \frac{e^{-kt}}t dt=-\sum c_k\log k$ by Frullani integrals. The integral of the other part of our sum $(*)$ is minus the limit at zero of the function $-\frac2{n+1}t^{-n-1}+\sum g_k(t^{-1})e^{-kt}$. The limit must exist, since the initial integral converges.

Now I cheat a bit. Note that if we write $f_n(t)=\sum_{k=0}^n q_k(1/t) e^{-kt}$ for polynomials $q_k$, then $c_k$ equals to the residue of the $q_k(1/t) e^{-kt}$, which is pretty computable. If I am not mistaken, $$q_k(t^{-1})=(-1)^kt^{-n-2}\left(2\binom{n-1}k+\binom{n-1}{k-1}(2+2t+t^2)\right)=\\=(-1)^kt^{-n-2}\left(2\binom{n}k+\binom{n-1}{k-1}(2t+t^2)\right).$$ Thus $$c_k=2(-1)^{k+n+1}\binom{n}k\frac{k^{n+1}}{(n+1)!}+2(-1)^{k+n}\binom{n-1}{k-1}\frac{k^{n}}{n!}+(-1)^{k+n+1}\binom{n-1}{k-1}\frac{k^{n-1}}{(n-1)!}=\\ =(-1)^{k+n+1}\frac{k^{n-1}(n^2+n-2k)}{n(n+1)(n-k)!(k-1)!}.$$ This matches a coefficient of $\log 3$ for $n=4$ from Shahrooz Janbaz's answer, you may check others for be sure. It remains to prove Sylvain JULIEN's guess for the rational part.

Ah, we may simply integrate by parts!

Denote $h(t)=(2 - e^{-t}( 2 + 2t+t^2))(1-e^{-t})^{n-1}$. Integrating by parts $(n+1)$ times we get $$\int_0^\infty h(t)t^{-n-2}dt=\frac1{(n+1)!}\int_0^\infty h^{(n+1)}(t)t^{-1}dt,$$ all off-integral terms vanish since $h(0)=h'(0)=\dots=h^{(n+1)}(0)=0$ and at infinity we have $h(t)=O(1)$, $h^{(i)}(t)=o(1)$ for $i>0$. We have $h^{(n+1)}(t)=\sum_{k=1}^n a_ke^{-kt}+u(t)$, where $u(t)=t\times \text{polynomial}(e^{-t},t)$. Note that $\sum a_k=h^{(n+1)}(0)=0$, thus $$\int_0^\infty \sum a_ke^{-kt}t^{-1}dt=\int_0^\infty \sum a_k(e^{-kt}-e^{-t})t^{-1}dt=-\sum a_k \log k$$ by Frullani integrals. It is easy to see that $$a_k=(-1)^{k+n+1}\binom{n-1}{k-1}k^{n-1}(n^2+n-2k).$$ It remains to evaluate $\int_0^\infty u(t) t^{-1}dt$. We have $$u(t)=-(2t+t^2)(g(t))^{(n+1)}-2(n+1)t(g(t))^{(n)},\,g(t)=e^{-t}(1-e^{-t})^{n-1}.$$ Therefore $$\int_0^\infty u(t) t^{-1}dt=-2\int g^{(n+1)}(t)dt-2(n+1)\int g^{(n)}(t)dt-\int_0^{\infty}tg^{(n+1)}(t)dt.$$ We get $\int (g^{(n)}+tg^{(n+1)})=tg^{(n)}$, and the definite integral against $(0,\infty)$ equals 0. It remains $\int_0^\infty -2g^{(n+1)}-(2n+1)g^{(n)}=2g^{(n)}(0)+(2n+1)g^{(n-1)}(0)$. We have $$g(t)=(1-t+\dots)t^{n-1}(1-t/2+\dots)^{n-1}=t^{n-1}-\frac{n+1}2t^n+\dots,$$ $g^{(n-1)}(0)=(n-1)!$, $g^{n}(0)=-\frac{(n+1)!}2$, $2g^{(n)}(0)+(2n+1)g^{(n-1)}(0)=-(n^2-n-1)(n-1)!$, confirming the guess of Sylvain JULIEN.

PREVIOUS NON COMPLETE VERSION

Still not a complete answer, but a method to be completed or improved.

Denote $f_n(t)=\frac{2 - e^{-t}\left ( 2 + 2t+t^2 \right )}{t^3}\left ( \frac{1-e^{-t}}{t} \right )^{n-1}$. Assume that we have found the coefficients $c_1,c_2,\dots,c_n$ and polynomials $g_1,\dots,g_n$ such that $$f_n(t)=2t^{-n-2}+\sum_{k=1}^n(g_k(t^{-1})e^{-kt})'+\sum_{k=1}^n c_k \frac{e^{-kt}}t.\,\,(*)$$ Then $\sum c_k=0$ (else $f_n$ would have a non-zero residue at 0, which is absurd). We have $\int_0^\infty \sum_{k=1}^n c_k \frac{e^{-kt}}t dt=-\sum c_k\log k$ by Frullani integrals. The integral of the other part of our sum $(*)$ is minus the limit at zero of the function $-\frac2{n+1}t^{-n-1}+\sum g_k(t^{-1})e^{-kt}$. The limit must exist, since the initial integral converges.

Now I cheat a bit. Note that if we write $f_n(t)=\sum_{k=0}^n q_k(1/t) e^{-kt}$ for polynomials $q_k$, then $c_k$ equals to the residue of the $q_k(1/t) e^{-kt}$, which is pretty computable. If I am not mistaken, $$q_k(t^{-1})=(-1)^kt^{-n-2}\left(2\binom{n-1}k+\binom{n-1}{k-1}(2+2t+t^2)\right)=\\=(-1)^kt^{-n-2}\left(2\binom{n}k+\binom{n-1}{k-1}(2t+t^2)\right).$$ Thus $$c_k=2(-1)^{k+n+1}\binom{n}k\frac{k^{n+1}}{(n+1)!}+2(-1)^{k+n}\binom{n-1}{k-1}\frac{k^{n}}{n!}+(-1)^{k+n+1}\binom{n-1}{k-1}\frac{k^{n-1}}{(n-1)!}=\\ =(-1)^{k+n+1}\frac{k^{n-1}(n^2+n-2k)}{n(n+1)(n-k)!(k-1)!}.$$ This matches a coefficient of $\log 3$ for $n=4$ from Shahrooz Janbaz's answer, you may check others for be sure. It remains to prove Sylvain JULIEN's guess for the rational part.

added 27 characters in body
Source Link
Fedor Petrov
  • 108.9k
  • 9
  • 264
  • 459

Still not a complete answer, but a method to be completed or improved.

Denote $f_n(t)=\frac{2 - e^{-t}\left ( 2 + 2t+t^2 \right )}{t^3}\left ( \frac{1-e^{-t}}{t} \right )^{n-1}$. Assume that we have found the coefficients $c_1,c_2,\dots,c_n$ and polynomials $g_1,\dots,g_n$ such that $$f_n(t)=2t^{-n-2}+\sum_{k=1}^n(g_k(t^{-1})e^{-kt})'+\sum_{k=1}^n c_k \frac{e^{-kt}}t.\,\,(*)$$ Then $\sum c_k=0$ (else $f_n$ would have a non-zero residue at 0, which is absurd). We have $\int_0^\infty \sum_{k=1}^n c_k \frac{e^{-kt}}t dt=-\sum c_k\log k$ by Frullani integrals. The integral of the other part of our sum $(*)$ is minus the limit at zero of the function $-\frac2{n+1}t^{-n-1}+\sum g_k(t^{-1})e^{-kt}$. The limit must exist, since the initial integral converges.

Now I cheat a bit. Note that if we write $f_n(t)=\sum_{k=0}^n q_k(1/t) e^{-kt}$ for polynomials $q_k$, then $c_k$ equals to the residue of the $q_k(1/t) e^{-kt}$, which is pretty computable (if. If I am not mistaken, $q_k(t^{-1})=(-1)^kt^{-n-2}(2\binom{n-1}k+\binom{n-1}{k-1}(2+2t+t^2))=(-1)^kt^{-n-2}(2\binom{n}k+\binom{n-1}{k-1}(2t+t^2))$.$$q_k(t^{-1})=(-1)^kt^{-n-2}\left(2\binom{n-1}k+\binom{n-1}{k-1}(2+2t+t^2)\right)=\\=(-1)^kt^{-n-2}\left(2\binom{n}k+\binom{n-1}{k-1}(2t+t^2)\right).$$ Thus $$c_k=2(-1)^{k+n+1}\binom{n}k\frac{k^{n+1}}{(n+1)!}+2(-1)^{k+n}\binom{n-1}{k-1}\frac{k^{n}}{n!}+(-1)^{k+n+1}\binom{n-1}{k-1}\frac{k^{n-1}}{(n-1)!}=\\ =(-1)^{k+n+1}\frac{k^{n-1}(n^2+n-2k)}{n(n+1)(n-k)!(k-1)!}.$$ This matches a coefficient of $\log 3$ for $n=4$ from Shahrooz Janbaz's answer, you may check others for be sure. It remains to prove Sylvain JULIEN's guess for the rational part.

Still not a complete answer, but a method to be completed or improved.

Denote $f_n(t)=\frac{2 - e^{-t}\left ( 2 + 2t+t^2 \right )}{t^3}\left ( \frac{1-e^{-t}}{t} \right )^{n-1}$. Assume that we have found the coefficients $c_1,c_2,\dots,c_n$ and polynomials $g_1,\dots,g_n$ such that $$f_n(t)=2t^{-n-2}+\sum_{k=1}^n(g_k(t^{-1})e^{-kt})'+\sum_{k=1}^n c_k \frac{e^{-kt}}t.\,\,(*)$$ Then $\sum c_k=0$ (else $f_n$ would have a non-zero residue at 0, which is absurd). We have $\int_0^\infty \sum_{k=1}^n c_k \frac{e^{-kt}}t dt=-\sum c_k\log k$ by Frullani integrals. The integral of the other part of our sum $(*)$ is minus the limit at zero of the function $-\frac2{n+1}t^{-n-1}+\sum g_k(t^{-1})e^{-kt}$. The limit must exist, since the initial integral converges.

Now I cheat a bit. Note that if we write $f_n(t)=\sum_{k=0}^n q_k(1/t) e^{-kt}$ for polynomials $q_k$, then $c_k$ equals to the residue of the $q_k(1/t) e^{-kt}$, which is pretty computable (if I am not mistaken, $q_k(t^{-1})=(-1)^kt^{-n-2}(2\binom{n-1}k+\binom{n-1}{k-1}(2+2t+t^2))=(-1)^kt^{-n-2}(2\binom{n}k+\binom{n-1}{k-1}(2t+t^2))$. Thus $$c_k=2(-1)^{k+n+1}\binom{n}k\frac{k^{n+1}}{(n+1)!}+2(-1)^{k+n}\binom{n-1}{k-1}\frac{k^{n}}{n!}+(-1)^{k+n+1}\binom{n-1}{k-1}\frac{k^{n-1}}{(n-1)!}=\\ =(-1)^{k+n+1}\frac{k^{n-1}(n^2+n-2k)}{n(n+1)(n-k)!(k-1)!}.$$ This matches a coefficient of $\log 3$ for $n=4$ from Shahrooz Janbaz's answer, you may check others for be sure. It remains to prove Sylvain JULIEN's guess for the rational part.

Still not a complete answer, but a method to be completed or improved.

Denote $f_n(t)=\frac{2 - e^{-t}\left ( 2 + 2t+t^2 \right )}{t^3}\left ( \frac{1-e^{-t}}{t} \right )^{n-1}$. Assume that we have found the coefficients $c_1,c_2,\dots,c_n$ and polynomials $g_1,\dots,g_n$ such that $$f_n(t)=2t^{-n-2}+\sum_{k=1}^n(g_k(t^{-1})e^{-kt})'+\sum_{k=1}^n c_k \frac{e^{-kt}}t.\,\,(*)$$ Then $\sum c_k=0$ (else $f_n$ would have a non-zero residue at 0, which is absurd). We have $\int_0^\infty \sum_{k=1}^n c_k \frac{e^{-kt}}t dt=-\sum c_k\log k$ by Frullani integrals. The integral of the other part of our sum $(*)$ is minus the limit at zero of the function $-\frac2{n+1}t^{-n-1}+\sum g_k(t^{-1})e^{-kt}$. The limit must exist, since the initial integral converges.

Now I cheat a bit. Note that if we write $f_n(t)=\sum_{k=0}^n q_k(1/t) e^{-kt}$ for polynomials $q_k$, then $c_k$ equals to the residue of the $q_k(1/t) e^{-kt}$, which is pretty computable. If I am not mistaken, $$q_k(t^{-1})=(-1)^kt^{-n-2}\left(2\binom{n-1}k+\binom{n-1}{k-1}(2+2t+t^2)\right)=\\=(-1)^kt^{-n-2}\left(2\binom{n}k+\binom{n-1}{k-1}(2t+t^2)\right).$$ Thus $$c_k=2(-1)^{k+n+1}\binom{n}k\frac{k^{n+1}}{(n+1)!}+2(-1)^{k+n}\binom{n-1}{k-1}\frac{k^{n}}{n!}+(-1)^{k+n+1}\binom{n-1}{k-1}\frac{k^{n-1}}{(n-1)!}=\\ =(-1)^{k+n+1}\frac{k^{n-1}(n^2+n-2k)}{n(n+1)(n-k)!(k-1)!}.$$ This matches a coefficient of $\log 3$ for $n=4$ from Shahrooz Janbaz's answer, you may check others for be sure. It remains to prove Sylvain JULIEN's guess for the rational part.

added 300 characters in body
Source Link
Fedor Petrov
  • 108.9k
  • 9
  • 264
  • 459

Still not a complete answer, but a method to be completed or improved.

Denote $f_n(t)=\frac{2 - e^{-t}\left ( 2 + 2t+t^2 \right )}{t^3}\left ( \frac{1-e^{-t}}{t} \right )^{n-1}$. Assume that we have found the coefficients $c_1,c_2,\dots,c_n$ and polynomials $g_1,\dots,g_n$ such that $$f_n(t)=2t^{-n-2}+\sum_{k=1}^n(g_k(t^{-1})e^{-kt})'+\sum_{k=1}^n c_k \frac{e^{-kt}}t.\,\,(*)$$ Then $\sum c_k=0$ (else $f_n$ would have a non-zero residue at 0, which is absurd). We have $\int_0^\infty \sum_{k=1}^n c_k \frac{e^{-kt}}t dt=-\sum c_k\log k$ by Frullani integrals. The integral of the other part of our sum $(*)$ is minus the limit at zero of the function $-\frac2{n+1}t^{-n-1}+\sum g_k(t^{-1})e^{-kt}$. The limit must exist, since the initial integral converges. For fixed $k$,

Now I cheat a bit. Note that if we get an equationwrite $f_n(t)=\sum_{k=0}^n q_k(1/t) e^{-kt}$ for polynomials $g_k$ and$q_k$, then $c_k$: $-kg_k(y)-y^2g_k'(y)+c_ky=q_k(y)$ for certain explicit polynomial equals to the residue of the $q_k$$q_k(1/t) e^{-kt}$, which is pretty computable (I denoteif I am not mistaken, $y=1/t$ here)$q_k(t^{-1})=(-1)^kt^{-n-2}(2\binom{n-1}k+\binom{n-1}{k-1}(2+2t+t^2))=(-1)^kt^{-n-2}(2\binom{n}k+\binom{n-1}{k-1}(2t+t^2))$. Note that thisThus $q_k$ has not too many monomials$$c_k=2(-1)^{k+n+1}\binom{n}k\frac{k^{n+1}}{(n+1)!}+2(-1)^{k+n}\binom{n-1}{k-1}\frac{k^{n}}{n!}+(-1)^{k+n+1}\binom{n-1}{k-1}\frac{k^{n-1}}{(n-1)!}=\\ =(-1)^{k+n+1}\frac{k^{n-1}(n^2+n-2k)}{n(n+1)(n-k)!(k-1)!}.$$ This matches a coefficient of $\log 3$ for $n=4$ from Shahrooz Janbaz's answer, about 3 or 4. This equation has unique solution whichyou may check others for be found by equating coefficientssure. It appears It remains to collect everything and get an answer. Note thatprove Sylvain JULIEN's guess for the answer we need only the constant term of the Laurent series $g_k(t^{-1})e^{-kt}$ and the value of $c_k$rational part.

Still not a complete answer, but a method to be completed or improved.

Denote $f_n(t)=\frac{2 - e^{-t}\left ( 2 + 2t+t^2 \right )}{t^3}\left ( \frac{1-e^{-t}}{t} \right )^{n-1}$. Assume that we have found the coefficients $c_1,c_2,\dots,c_n$ and polynomials $g_1,\dots,g_n$ such that $$f_n(t)=2t^{-n-2}+\sum_{k=1}^n(g_k(t^{-1})e^{-kt})'+\sum_{k=1}^n c_k \frac{e^{-kt}}t.\,\,(*)$$ Then $\sum c_k=0$ (else $f_n$ would have a non-zero residue at 0, which is absurd). We have $\int_0^\infty \sum_{k=1}^n c_k \frac{e^{-kt}}t dt=-\sum c_k\log k$ by Frullani integrals. The integral of the other part of our sum $(*)$ is minus the limit at zero of the function $-\frac2{n+1}t^{-n-1}+\sum g_k(t^{-1})e^{-kt}$. The limit must exist, since the initial integral converges. For fixed $k$, we get an equation for $g_k$ and $c_k$: $-kg_k(y)-y^2g_k'(y)+c_ky=q_k(y)$ for certain explicit polynomial $q_k$ (I denote $y=1/t$ here). Note that this $q_k$ has not too many monomials, about 3 or 4. This equation has unique solution which may be found by equating coefficients. It appears to collect everything and get an answer. Note that for the answer we need only the constant term of the Laurent series $g_k(t^{-1})e^{-kt}$ and the value of $c_k$.

Still not a complete answer, but a method to be completed or improved.

Denote $f_n(t)=\frac{2 - e^{-t}\left ( 2 + 2t+t^2 \right )}{t^3}\left ( \frac{1-e^{-t}}{t} \right )^{n-1}$. Assume that we have found the coefficients $c_1,c_2,\dots,c_n$ and polynomials $g_1,\dots,g_n$ such that $$f_n(t)=2t^{-n-2}+\sum_{k=1}^n(g_k(t^{-1})e^{-kt})'+\sum_{k=1}^n c_k \frac{e^{-kt}}t.\,\,(*)$$ Then $\sum c_k=0$ (else $f_n$ would have a non-zero residue at 0, which is absurd). We have $\int_0^\infty \sum_{k=1}^n c_k \frac{e^{-kt}}t dt=-\sum c_k\log k$ by Frullani integrals. The integral of the other part of our sum $(*)$ is minus the limit at zero of the function $-\frac2{n+1}t^{-n-1}+\sum g_k(t^{-1})e^{-kt}$. The limit must exist, since the initial integral converges.

Now I cheat a bit. Note that if we write $f_n(t)=\sum_{k=0}^n q_k(1/t) e^{-kt}$ for polynomials $q_k$, then $c_k$ equals to the residue of the $q_k(1/t) e^{-kt}$, which is pretty computable (if I am not mistaken, $q_k(t^{-1})=(-1)^kt^{-n-2}(2\binom{n-1}k+\binom{n-1}{k-1}(2+2t+t^2))=(-1)^kt^{-n-2}(2\binom{n}k+\binom{n-1}{k-1}(2t+t^2))$. Thus $$c_k=2(-1)^{k+n+1}\binom{n}k\frac{k^{n+1}}{(n+1)!}+2(-1)^{k+n}\binom{n-1}{k-1}\frac{k^{n}}{n!}+(-1)^{k+n+1}\binom{n-1}{k-1}\frac{k^{n-1}}{(n-1)!}=\\ =(-1)^{k+n+1}\frac{k^{n-1}(n^2+n-2k)}{n(n+1)(n-k)!(k-1)!}.$$ This matches a coefficient of $\log 3$ for $n=4$ from Shahrooz Janbaz's answer, you may check others for be sure. It remains to prove Sylvain JULIEN's guess for the rational part.

Source Link
Fedor Petrov
  • 108.9k
  • 9
  • 264
  • 459
Loading