10
$\begingroup$

Let $\alpha \vdash d$ be a partition of $d$, i.e. $\alpha = (\alpha_1 \geq \alpha_2 \geq …\geq \alpha_l)$, where $\sum_k \alpha_k = d$. Define a Laurent polynomial in $Q$ as follows:

$$ P_\alpha(Q) = \sum_{k=1}^\infty Q^{-\alpha_k + k-1}(1-Q) \\ \quad \quad \quad \quad \quad \quad = Q^{l} + \sum_{k=1}^{l} (Q^{-\alpha_k + k-1} - Q^{-\alpha_k + k}) $$

Question 1: Has anyone encountered such polynomials and know a nice formula for them and/or a reference to a place where they have appeared?

I can prove that $P_\alpha(Q) = P_{\alpha'}(Q^{-1})$ where $\alpha'$ is the conjugate partition.

The sum of these polynomials over all $\alpha \vdash d$ appears to be nice but I have not discovered a good closed formula. I would especially like a formula for the coefficients of the following generating series in $v$, which has arisen in a computation in Donaldson-Thomas theory. Define Laurent polynomials $C_d(Q)$ by the following formula:

$$ \sum_{d=0}^\infty C_d (Q)\,\, v^d = \frac{Q}{(1-Q)^2}\left( \sum_\alpha v^{|\alpha|} P_\alpha \right) \prod_{m=1}^\infty (1-v^m). $$

(Note that $C_0$ is not a polynomial, but for $d>0$, $C_d$ seems to be a polynomial). Computation for $d$ up to 40 suggests that

$C_d(Q)$ is some sort of $Q$ deformation of $\sigma(d)$, the sum of divisors function.

Question 2: Prove that $C_d(1)=\sigma(d)$. Is there a formula for $C_d(Q)$ which looks like a $Q$-deformation of the formula $\sigma(d) = \sum_{k|d}k$? Or is there some other natural explanation for this phenomenon?

Below is a list of $C_d$ for $d$ up to 9:

$$C_1 = 1$$ $$C_2 = Q+1+1/Q$$ $$C_3 = Q^2+Q+1/Q+1/Q^2$$ $$C_4 = Q^3+Q^2+Q+1+1/Q+1/Q^2+1/Q^3$$ $$C_5 = Q^4+Q^3+Q^2+1/Q^2+1/Q^3+1/Q^4$$ $$C_6 = Q^5+Q^4+Q^3+Q^2+Q+2+1/Q+1/Q^2+1/Q^3+1/Q^4+1/Q^5$$ $$C_7 = Q^6+Q^5+Q^4+Q^3+1/Q^3+1/Q^4+1/Q^5+1/Q^6$$ $$C_8 = Q^7+Q^6+Q^5+Q^4+Q^3+Q^2+Q+1+1/Q+1/Q^2+1/Q^3+1/Q^4+1/Q^5+1/Q^6+1/Q^7$$ $$C_9 = Q^8+Q^7+Q^6+Q^5+Q^4+Q+1+1/Q+1/Q^4+1/Q^5+1/Q^6+1/Q^7+1/Q^8$$

$\endgroup$
  • $\begingroup$ "$P_\alpha\left(Q\right) = P_{\alpha^{\prime}}\left(Q^{-1}\right)$" -- I'd be careful with such formulas; is there a well-defined way to make sense of them? The two sides of the equation certainly don't match as two-sided power series; I assume you mean that some power of $1-Q$ kills their difference. (It would be interesting to know which one.) $\endgroup$ – darij grinberg May 5 '14 at 0:42
  • 1
    $\begingroup$ @darijgrinberg $P_\alpha$ is a Laurent polynomial so my equation makes sense. The infinite sum is really a finite sum since for large $k$ the sum telescopes. I'll edit the question to make that more clear. $\endgroup$ – Jim Bryan May 5 '14 at 2:53
  • $\begingroup$ Ah! Good point. $\endgroup$ – darij grinberg May 5 '14 at 4:40
7
$\begingroup$

Closely related infinite series appear in relation to the infinite wedge; this then gives a proof of your second question.

Warnings: I skimp out on the details of a slightly laborious Calculus I exercise at the end, but the answer was already getting long enough.

Also, it looks like I probably made a sign error, presumably related to $Q^{1/2}-Q^{-1/2}$ transforming under $Q\to 1/Q$ and how your invariants behave under conjugation.

A related infinite series

In the fermionic viewpoint for partitions, the natural thing to consider is not your Laurent polynomials, but the infinite series

$$f_\alpha(Q)=\sum_{k=1}^\infty Q^{\alpha_k-k+1/2}.$$

In the bijection between partitions and charge zero elements of Dirac's electron sea, the exponents here are exactly the energies of the electrons.

Your function simply differs by a factor: $P_\alpha(Q^{-1})=(Q^{1/2}-Q^{-1/2})f_\alpha(Q)$.

The $f_\alpha$ occur naturally as eigenvalues of operators on the infinite wedge; in the notation of Okounkov and Pandharipande's trilogoy on the Gromov-Witten theory of curves, for instance, we have: $$\mathcal{E}_0(\ln(Q))v_\lambda=f_\lambda(Q)v_\lambda$$

But for your second question, the source you want is Bloch and Okounkov The character of the infinite wedge.

A result of Bloch and Okounkov

The pertinent bit to you is the easiest example of their Theorem 0.5, which they prove as Theorem 6.5 as a warm-up to the main event.

Some notation: for a function on polynomials, they define $$\langle f\rangle_q=\sum_\lambda f(\lambda)q^{|\lambda|}/\sum_{\lambda} q^{|\lambda|}=\left(\sum_\lambda f(\lambda)q^\lambda\right)\prod (1-q^n),$$the expected value of $f$ over partitions if a partition $\lambda$ has probability $q^{|\lambda|}$.

Then, Theorem 6.5 of Bloch-Okounkov gives: $$\left\langle \sum q^{\alpha_k-k+1/2}\right\rangle_q=\frac{1}{\Theta(t)}$$ where $$\Theta(t)=\frac{(t^{1/2}-t^{-1/2})(qt)_\infty (q/t)_\infty}{(q)_\infty^2}$$ with the standard Pochhammer symbol $(a)_\infty=\prod_{n\geq 0} (1-aq^n)$.

Product formula for $\sum C_d(Q)v^d$

Bloch and Okounkov's result immediately gives an infinite product formula for your $\sum C_d(Q)v^d$. I am going to change your notation to be consistent with Bloch-Okounkov and Pochhammer symbols -- $Q$ becomes $t$, and $v$ becomes $q$.

The right hand side of your formula for $\sum C_d(t) q^d$ becomes

$$\frac{t}{(1-t)^2} \left\langle (t^{1/2}-t^{-1/2})\sum t^{\alpha_k-k+1/2}\right\rangle_q $$

We can pull the $(t^{1/2}-t^{-1/2})$ outside the expectation, and simplify the resulting factors involving $t$ to $1/(t^{1/2}-t^{-1/2})$. Then we can substitute Bloch-Okounkov's result, to get

$$\sum C_d(t)q^d=\frac{(q)_\infty^2}{(t^{1/2}-t^{-1/2})^2(qt)_\infty (q/t)_\infty}$$

A $t$-deformation of $\sigma(d)$

We now show that this implies $C_d(1)=\sigma(d)$.

We would like to take the limit as $t\to 1$ of our product formula, but this limit does not exist because the constant in $q$ term is $$\frac{1}{(1-t)(1-t^{-1})}$$ (I think this is right, and there was a sign error in a previous formula). We subtract this term out by hand to remove the pole, and then take the limit:

$$\lim_{t\to 1} \frac{(q)_\infty^2-(qt)_\infty(q/t)_\infty}{(1-t)(1-1/t)(qt)_\infty(q/t)_\infty}$$

Note that the limit approaches $0/0$, and using l'Hôpital's rule once and simplifying, one sees that this limit is indeed

$$\sum_{n\geq 1}\frac{q^n}{(1-q^n)^2}=\sum_{n\geq 1} \sigma(n)q^n.$$

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ Awesome answer! Thanks, Paul. That exactly what I needed and wanted. BTW, there are a couple typos which you might want to edit for posterity's sake :-). (1) "for a function on partitions". (2) Inside the bracket on the LHS of the equation after "Bloch-Okounkov gives" the $q$ should be a $t$. $\endgroup$ – Jim Bryan May 8 '14 at 14:54

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.