I was doing some sums when this idea popped into my head. What is the $$\sum_{i=1}^n i^i$$ I have been trying to find a relation using induction but hadn't had any succes. Any other ideas? What about this other one? $$\sum_{i = 1} ^ n i ^ {1/i}$$
-
9It's very likely that these sums don't have closed-form expressions, let alone elementary ones. – Alann Rosas Jul 01 '21 at 19:35
-
There is a partial solution in the preview image for bounds on the first sum here – user170231 Jul 01 '21 at 19:36
-
I think the second sum is asymptotically equivalent to $n$ (in the sense that $\frac{1}{n} \sum_{i=1}^n i^{1/i} \to 1$ as $n \to \infty$). I don't expect there to be a nice expression for the exact value though. – angryavian Jul 01 '21 at 19:38
-
google it. That is, google the first few terms, for the first sequence google 1 5 29 259 3129 oeis. The result is sequence https://oeis.org/A231712 which has common term $n^n+n-1$. That is, $\sum_{i=1}^n i^i=n^n+n-1$ – Mirko Jul 01 '21 at 19:56
-
@Mirko I didn't know about this way to find sequences. Thanks! – andu eu Jul 01 '21 at 20:02
-
2@Mirko the formula you provided does not agree with the sum for $n\geq 3$. I'm pretty sure it's because the sequence terms you fed to Google are incorrect; they should be $1$, $5$, $32$, $288$, and $3413$, respectively. That said, numerics suggest that your formula and the sum are asymptotically equivalent, that is, $\lim_{n\to\infty}\frac{1}{n^n+n-1}\sum_{i=1}^n i^i =1$. – Alann Rosas Jul 01 '21 at 20:16
-
@Alann You are right (I computed the terms using Excel and made a typo, A2^A2+A1 instead of the correct one A2^A2+B1). So the first sequence is https://oeis.org/A001923 and indeed, even if one could find interesting info there (and relations to other sequences), it doesn't look like it has a known closed-form expression. – Mirko Jul 01 '21 at 20:27
-
Perhaps it's interesting to find an asymptotic formula of the two series for $n \to +\infty$ – NN2 Jul 01 '21 at 21:32
1 Answers
Asymptotic expansions for these sums can be obtained as follows.
Consider the first sum. We can write $$ \sum\limits_{i = 1}^n {i^i } = \sum\limits_{i = 0}^{n - 1} {(n - i)^{n - i} } = n^n \sum\limits_{i = 0}^{n - 1} {\frac{1}{{n^i }}\left( {1 - \frac{i}{n}} \right)^{n - i} } . $$ Now $$ \left( {1 - \frac{i}{n}} \right)^{n - i} = \exp \left( {n\log \left( {1 - \frac{i}{n}} \right) - i\log \left( {1 - \frac{i}{n}} \right)} \right) $$ and, by Taylor expansion, $$ \log \left( {1 - \frac{i}{n}} \right) = - \sum\limits_{k = 1}^\infty {\frac{{i^k }}{k}\frac{1}{{n^k }}} . $$ Substituting the latter into the former and expanding the exponential, we eventually find $$ \left( {1 - \frac{i}{n}} \right)^{n - i} = e^{ - 1} \left( {1 + \frac{{i^2 }}{2}\frac{1}{n} + \frac{{i^3 (3i + 4)}}{{24}}\frac{1}{{n^2 }} + \cdots } \right). $$ Substituting these expansions into the expression for $\sum_{i=1}^n i^i$ and re-arranging, we deduce $$\boxed{ \sum\limits_{i = 1}^n {i^i } \sim n^n \left( {1 + \frac{1}{e}\frac{1}{n} + \left( {\frac{1}{{2e}} + \frac{1}{{e^2 }}} \right)\frac{1}{{n^2 }} + \left( {\frac{7}{{24e}} + \frac{2}{{e^2 }} + \frac{1}{{e^3 }}} \right)\frac{1}{{n^3 }} + \cdots } \right)} $$ as $n\to +\infty$. The general term may be written $P_k (e^{ - 1} )n^{-k}$, where $P_k$ is a polynomial of degree $k$. A more detailed analysis shows that $$ P_k (x) = \frac{1}{{k!}}\sum\limits_{j = 0}^k {\left[ {\frac{{d^k }}{{dt^k }}\left( {t^j \exp \left( {\sum\limits_{p = 1}^{k - j} {\frac{{j^{p + 1} }}{{p(p + 1)}}t^p } } \right)} \right)} \right]_{t = 0} x^j } . $$
Consider now the second sum. We can write $$ \sum\limits_{i = 1}^n {i^{1/i} } = n + \sum\limits_{i = 1}^n {\frac{{\log i}}{i}} + \sum\limits_{i = 1}^n {\frac{1}{2}\left( {\frac{{\log i}}{i}} \right)^2 } + \sum\limits_{i = 1}^n {\left( {i^{1/i} - 1 - \frac{{\log i}}{i} - \frac{1}{2}\left( {\frac{{\log i}}{i}} \right)^2 } \right)} . $$ By the Euler–Maclaurin formula, $$ \sum\limits_{i = 1}^n {\frac{{\log i}}{i}} = \frac{{\log ^2 n}}{2} + \frac{{\log n}}{{2n}} + a + \mathcal{O}\!\left( {\frac{{\log n}}{{n^2 }}} \right), $$ $$ \sum\limits_{i = 1}^n {\frac{1}{2}\left( {\frac{{\log i}}{i}} \right)^2 } = - \frac{{\log ^2 n + 2\log n + 2}}{2n} + b + \mathcal{O}\!\left( {\frac{{\log ^2 n}}{{n^2 }}} \right) $$ and $$ \sum\limits_{i = 1}^n {\left( {i^{1/i} - 1 - \frac{{\log i}}{i} - \frac{1}{2}\left( {\frac{{\log i}}{i}} \right)^2 } \right)} = c + \mathcal{O}\!\left( {\frac{{\log ^3 n}}{{n^2 }}} \right) $$ with some constants $a$, $b$, and $c$. Consequently, $$\boxed{ \sum\limits_{i = 1}^n {i^{1/i} } = n + \frac{{\log ^2 n}}{2} + \kappa - \frac{{\log ^2 n + \log n + 2}}{{2n}}+\mathcal{O}\!\left( {\frac{{\log ^3 n}}{{n^2 }}} \right)} $$ as $n\to +\infty$, with some constant $\kappa$. Solving for $\kappa$ and taking $n$ large, it is found that $\kappa = 0.988549601142269\ldots$. With more work, one can derive an asymptotic expansion of the form $$ \sum\limits_{i = 1}^n {i^{1/i} } \sim n + \frac{{\log ^2 n}}{2} + \kappa + \sum\limits_{k = 1}^\infty {\frac{{Q_{k + 1} (\log n)}}{{n^k }}} $$ where $Q_k$ is a polynomial of degree $k$. I did not make an attempt to obtain a formula for these polynomials though.

- 31,845