EDIT: This is a case of being too wrapped up in a formulation ($e_j,p_i,$ and the like) to try something simple. It did not occur to me to pull exp to the outside in the weeks I have stared at this. Thanks to Aaron Meyerowitz and Brendan McKay for a humbling revelation. I leave it to others to decide if this is an object lesson in mathematical myopia or a question to be removed. END EDIT

If the following turns out to be a homework question, great! Just tell me what course, textbook (or online accessible .pdf) and page number has the problem, and I'll go check it out.

Let $m$ and $k$ be integers with $k \gt e^m \geq 1$. I am considering two expressions in variables $x_i$, and looking at the partial sums with monomials of degree at most $m$, which I note by $\cong_m$. (So if any term like, say, $x_hx_i^{m-1}x_j$ occurs in my sum, I toss that term and consider what is left with $\cong_m$.) The question in brief is: $$\prod_{0 \leq i \leq k}(1 - x_i) \cong_m \prod_{1 \leq j}(\sum_{0 \leq l} [ - \sum_{0 \leq i \leq k}(x_i)^j/j ]^l/l! ) ?$$

Here is some background and motivation. I rewrite the above question in terms of elementary symmetric polynomials in the $x_i, 0\leq i \leq k$.
In order to have enough terms to make sense, assume $k$ is as large as needed. (Likely I just need $k > e^{e^m}$, but it would be nice to know if, say, just $k \gt m^2$ will do.) Let us write $q_j$ for $ \sum_{0 \leq i \leq k} (x_i)^j/j $. This is $p_j/j$, or the $jth$ power symmetric polynomial divided by $j$. Let us also write $XP_m(t)$ for some polynomial $t$ to be the formal expression (which we will chop off after the terms get too big) $\sum_{0\leq j \leq m} (-t)^j/j!$, after the series expansion for $e^{-t}$. For the left hand side, we rewrite it in terms of the elementary symmetric polynomials $e_j$, which is the sum over all $j$-sets (not $j$-tuples) $\{i_1, \ldots , i_j\}$ of $\{0,\ldots,k\}$ of the monomials $x_{i_1}\ldots x_{i_j}$. I also declare $e_0=1$. Now the left hand side is written as a sum, and as I am concerned only with that part containing monomials of total degree at most $m$, I can limit the summation indices and the question now becomes:

$$\sum_{ 0 \leq i \leq m} (-1)^ie_i \cong_m \prod_{1\leq j \leq m} XP_m(q_j) ?$$

If I had facility with any computer algebra system, I would have tried it there first before asking here. I have verified the equality for small $m$, and would like a reference to a proof, or disproof. Also, Newton's identities $je_j = \sum_{1\leq i \leq j}(-1)^{i-1}p_ie_{j-i}$ give me some hope that the above is true, but I have not found a derivation.

My motivation is in exploring the limits of an argument I learned in a paper of Harlan Stevens. He uses the left hand side with prime reciprocal values for the $x_i$, and looks for the smallest odd integer m such that the sum is positive, but all partial sums for smaller odd positive integers are negative, so in particular $q_1=p_1\gt 1$ for interesting examples.
If it is true, I hope to show that the smallest odd positive $m$ that gives a positive partial sum implies $p_1 \gt Cm$ for some real constant $C$, which would then tell me that this argument is limited to showing upper bounds no tighter than $Ak^{B\log\log k}$, and no smaller.

Gerhard "Also, I Find It Pretty" Paseman, 2014.01.19

  • $\begingroup$ It occurs to me that this might be in a text like Enumerative Combinatorics or generatingfunctionality. If someone knows where this is treated, a text and section number would be appreciated. Gerhard "Lacks Good Web Search Terms" Paseman, 2014.01.19 $\endgroup$ Jan 19, 2014 at 22:49
  • $\begingroup$ DO you know that $k=m$ is not enough? $\endgroup$ Jan 19, 2014 at 23:16
  • $\begingroup$ Gerhard, unclear, do you have the two books you mention? I think I have both as pdfs. I found the second one already.. $\endgroup$
    – Will Jagy
    Jan 20, 2014 at 0:05
  • $\begingroup$ Your first display has "[" once and "]" twice. Please fix it. $\endgroup$ Jan 20, 2014 at 1:04
  • 1
    $\begingroup$ I don't really get the question. The right side of your first display is just a rearrangement of $\exp(\sum_i \ln(1-x_i))$ after applying Taylor expansion to the exp and ln functions. When it is formally expanded out, all terms with total degree more than $k$ vanish, as they should by Taylor's theorem. $\endgroup$ Jan 20, 2014 at 1:21

1 Answer 1


Try this: $$\prod_{1 \leq j}^{\infty}(\sum_{0 \leq l}^{\infty} [ - \sum_{i=0}^k\frac{x_i^j}{j} ]^l/l! ])=$$ $$\prod_{1 \leq j}^{\infty}\mathop{exp}\left( - \sum_{i=0}^k\frac{x_i^j}{j}\right)=$$ $$\mathop{exp}\left(\sum_{j=1}^{\infty}\left( - \sum_{i=0}^k\frac{x_i^j}{j}\right)\right)=$$ $$\mathop{exp}\left(\sum_{i=0}^k\left( -\sum_{j=1}^{\infty} \frac{x_i^j}{j}\right)\right)=$$ $$\mathop{exp}\left(\sum_{i=0}^k\ln({1-x_i)}\right)=$$ $$=\prod_{i=0}^k(1-x_i )$$

So I think $m=k$ would be just enough.

  • $\begingroup$ Darn, beat by 3 seconds! $\endgroup$ Jan 20, 2014 at 1:21
  • $\begingroup$ Oh, it's only that Brendan types faster. Gerhard "You Still Get The Upvote" Paseman, 2014.01.20 $\endgroup$ Jan 20, 2014 at 22:41

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.