Edit. I added two references to the work of Miloš Kössler I found recently: he, by using the variable transformation $z\mapsto \zeta + \eta e^{-i\theta} \zeta^2$, $0< \eta \le {3/ 4}$, gives two necessary and sufficient condition for the power series \eqref{1} to have the point $e^{i\theta}$ as a singular boundary point. These condition, which do not rely on the one of Fabry described below, are briefly described in the final notes.
Terminology: by singular boundary point I mean a boundary point across which you cannot analytically continue an analytic function defined on the interior of the give domain, while this is always possible at a regular boundary point.
A necessary and sufficient condition.
I am not aware of any necessary condition for the "lacunarity". However, a conceptually simple necessary and sufficient condition for a boundary point to be a singular point is known: and as a corollary, you can use it and check every boundary point of the domain of your function is singular or not.
Without any restriction on generality, let's consider a power series
$$
f(z)=\sum_{n=0}^{\infty} a_n z^n,\label{1}\tag{1}
$$
whose radius of convergence $R_f$ is 1 and suppose we want to check if $z=1$ is singular or not.
If the expansion of $f(x)$ as a power series centered at any point on the real segment $]0,1[$ converges on a disk which includes $z=1$ then this point is regular, otherwise it is singular.
This form of the condition is fully given by Markushevich ([2] chapter IX §IX.7 pp313-314) and Titchmarsh ([3] chapter 7, §7.23, p. 216): however, this latter author follows Landau in simplifying the calculations by introducing the following function $F(\zeta)$ ([3] chapter 7, §7.23, pp. 216-217, [1] chapter 5, §19, p. 76-77). Let
$$
F(\zeta)=\frac{1}{1-\zeta}f\left(\frac{\zeta}{1-\zeta}\right)=\sum_{n=0}^{\infty} b_n z^n,
$$
where $b_n=\sum_{m=0}^{n}\binom{n}{m} a_m$. Then $z=1$ is a singular boundary point for $f$ if and only if
$$
R_F=\limsup_{n\to\infty}|b_n|^{-\frac{1}{n}}=\frac{1}{2}, \label{2}\tag{2}
$$
and thus we get the following
Corollary. Let $f:\Bbb C\to\Bbb C$ be an analytic function whose power series expansion at $0\in\Bbb C$ is \eqref{1}. Then $f$ is lacunary on the open unit disk $\Bbb D$ (i.e. $\partial\Bbb D$ is the "natural boundary" for $f$) if and only if
$$
\limsup_{n\to\infty} |a_n|^\frac{1}{n}=1\;\wedge\;\limsup_{n\to\infty}{\left|\sum_{m=0}^{n}\binom{n}{m} a_me^{im\theta}\right|^\frac{1}{n}}\!\!=2\quad\forall \theta\in[0,2\pi]
$$
Notes
- According to Landau ([1] chapter 5, §19, p. 76), this criterion is due to Fabry.
- Reference [2] have been translated in English as Markushevich A.I.
The theory of analytic functions:
a brief course, Moscow: MIR: however, I do not have access to a copy of that book, therefore I refer to the Italian edition listed below.
- Miloš Kössler, by using the variable transformation $z\mapsto \zeta + \eta e^{-i\theta} \zeta^2$, transforms \eqref{1} in the power series $F(\zeta)=\sum_{n=0}^{\infty} A_n(\eta, e^{i\theta}) \zeta^n$. Then proves that ([A1], p. 27) a necessary and sufficient condition for the point $e^{i\theta}$ to be a singular boundary point for \eqref{1} is that
$$
\limsup_{n \to \infty} {|A_n(\eta, e^{i\theta})|^{1\over n}}=\frac{2\eta}{\sqrt{1+4\eta}-1}\qquad 0< \eta \le {3\over 4}\label{3}\tag{2b}
$$
Obviously, if the above value is independent from $\theta\in [0,2\pi]$ then the all the boundary of the unit disk is made of singular points for \eqref{1}. Kössler's condition may be interesting due to the fact that while the coefficients $\{b_n\}_{n\in \Bbb N}$ above are customarily polynomials of order $n$ respect to $e^{i\theta}$, the coefficients $A_n$ are polinomials at most of degree $\big[{n \over2}\big]$ (respect to the same variable). In the paper [A2] (pp. 528-529), with the same title, he further simplifies condition \eqref{3} assuming directly $\eta= {3/4}$ and defining a sequence $\{B_n\}_{n\in \Bbb N}$ by which \eqref{3} becomes
$$
\limsup_{n \to \infty} {|B_n(e^{i\theta})|^{1\over n}}= {3\over 4}.\label{4}\tag{2c}
$$
Each $B_n$ is a polynomial whose (variable, depending on an arbitrarily small positive constant $\mu$) number of terms is howhever less than $\big[{n \over2}\big]$, thus again \eqref{4} could be easier to veryfy respect to conditions \eqref{2} and \eqref{3} for some power series $f$.
References
[1] Landau, Edmund; Gaier, Dieter, Darstellung und Begründung einiger neuerer Ergebnisse der Funktionentheorie. 3 erw. Auflage (German), Berlin-Heildelberg-New York: Springer-Verlag, pp. XI+201 (1986), ISBN: 3-540-16886-9, MR0869998, Zbl 0601.30001.
[2] Markushevich, Alekseĭ Ivanovich, Elementi di teoria delle funzioni analitiche. Translated from the Russian by Ernest Kozlov, (Italian)
Nuova Biblioteca di Cultura, Serie Scientifica. Roma: Editori Riuniti; Moscow: Edizioni Mir. pp. 384 (1988), ISBN: 88-359-3284-X, MR1011460, Zbl 0694.30002.
[3] Titchmarsh, Edward Charles, The theory of functions 2nd ed., (English), Oxford: Oxford University Press, pp. X+454 (1939), JFM 65.0302.01, MR3728294, Zbl 0336.30001.
Addendum reference
[A1] M. Kössler, "Sur les singularités des séries entières" (French), Atti della Reale Accademia dei Lincei, Rendiconti, Classe di Scienze Fisiche Matematiche Naturali, Serie V, 32, No. 1, pp. 26-29 (1923), JFM 49.0236.01.
[A2] M. Kössler, "Sur les singularités des séries entières" (French), Atti della Reale Accademia dei Lincei, Rendiconti, Classe di Scienze Fisiche Matematiche Naturali, Serie V, 32, No. 1, pp. 528-531 (1923), JFM 49.0236.03.
