10
$\begingroup$

One can use Lagrange inversion to find the power series $F(x)$, which solves $F(x) = x(1+F(x)^p)$, where $p$ is a positive integer.

Now, what if $p$ is not an integer, but rather a positive rational number, say $p=7/3$?

As a concrete example, we are looking for a formal solution to $F(x) = x(1+F(x)^{7/3})$, but now, $F(x) \in \mathbb{C}[x^{1/3}]$.

The Lagrange inversion formula still seem to work in this case, that is, the function $$ F(x) := \sum_{r>0} x^r \left( [t^{r-1}]\frac{1}{r} (1+t^p)^r \right) $$ is a solution to $F(x) = x(1+F(x)^p)$, but now we must have $F(x) \in \mathbb{C}[x^{1/d}]$, where $d$ is the denominator of $p$, and the sum ranges over all positive integer multiples of $1/d$.

Is there some reference which proves this extension of Lagrange inversion?

Edit: I think I managed to prove that Lagrange inversion generalizes to this setting, i.e, instead of having $f,g \in \mathbb{C}[x]$, we have $f,g \in \mathbb{C}[x^{1/d}]$, and we wish to express the coefficients of $g$, in terms of the coefficients of $f$, where $f(g(x))=x$.

$\endgroup$
13
  • $\begingroup$ but for $r=1$ we get $[x]F=1$, so $F=x+\ldots$ and $F^p$ is not a power series? $\endgroup$ Commented Jan 5, 2022 at 10:51
  • $\begingroup$ @FedorPetrov right, we need to allow for fractional powers also. I think actually I can manage to prove the above, so I am typing it up... the proof given in users.math.msu.edu/users/magyarp/math880/Lagrange.pdf should generalize to the setting i am interested in.. $\endgroup$ Commented Jan 5, 2022 at 11:16
  • $\begingroup$ But $F/x$ is a power series while $1+F^p$ is not, how can they be equal? $\endgroup$ Commented Jan 5, 2022 at 11:40
  • $\begingroup$ Well, $F(x)$ is not a power series, but rather a power-series in the variable $x^{1/d}$. $\endgroup$ Commented Jan 5, 2022 at 12:19
  • 2
    $\begingroup$ @PerAlexandersson In cases like this it's often useful to apply Lagrange inversion to an equation like $f = R(f)$ rather than $f=xR(f)$. Thus if we want to solve $F =x (1+F^p)$ we can set $F(x) = xG(x^p)$ and $y=x^p$; then $F=x(1+F^p)$ reduces to $G(y) = 1+ yG(y)^p$, the well-known Fuss-Catalan functional equation. $\endgroup$
    – Ira Gessel
    Commented Jan 5, 2022 at 19:12

2 Answers 2

9
$\begingroup$

Let $p$ be any non-zero complex exponent. Let $f \in x\big(1+\mathbb C[[x]]\big)\subset \mathbb C[[x]]$ be the well-known compositional inverse of $x(1+x)^{-p}\in x\big(1+\mathbb C[[x]]\big)$ given by the Lagrange inversion: $$f(x)=\sum_{n=1}^\infty\frac1n {pn\choose n-1}x^n,$$

and consider its formal conjugate $F(x):=f(x^p)^{1/p}$, thus an element of $M:=x\big(1+\mathbb C[[x^p]]\big)$. Since $f$ solves $f(x)=x\big(1+f(x)\big)^p$ we have that $F$ solves in $M$ $$F(x)=x\big(1+F(x)^p\big).$$ This on the formal side; also, e.g. for small real nonnegative $x$, one gets a convergent series.

(Not needed here). There is also an easy extension of the Lagrange inversion formula that already works in $X^\alpha\mathbb C((X))$ as a $\mathbb C((X))$-module, for $\alpha\in\mathbb C$. I reported it in the wiki article. An application is e.g. the power series expansion of the complex powers of the Lambert function.

$\endgroup$
1
  • $\begingroup$ Ah, that's nice. $\endgroup$ Commented Jan 5, 2022 at 14:05
4
$\begingroup$

I managed to write up a proof based on these lecture notes (which cites Stanley's EC2)

The original proof uses residue calculus, and properties of analytic functions. This proof does not generalize to rational exponents, but the proof linked above, extends without any issues to "power series" with rational powers.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .