The Severi inequality \(K^2 \geq 4\chi\) for surfaces of maximal Albanese dimension (Q1770273)

The main result of this beautiful paper is the proof of the Severi inequality, that states that if a minimal surface of general type has maximal Albanese dimension (i.e. if the image of the Albanese morphism is a surface), then \(K^2_S \geq 4\chi\). The inequality takes its name from F. Severi since he was the first to claim the result in the 30's [\textit{F. Severi}, Comment. Math. Helv. 4, 268--326 (1932; Zbl 0005.17602)]. His proof was wrong, as pointed out by \textit{F. Catanese} [in: Algebraic geometry-open problems, Proc. Conf. Ravello/Italy 1982, Lect. Notes Math. 997, 90--112 (1982; Zbl 0517.14011)], since based on the assertion that a surface with irrationality \(q\) either contains an irrational genus \(q\) fibration, or \(H^0(\Omega^1_S)\) has no base points. Counterexamples to this assertion were given by \textit{F. Catanese} [J. Differential Geom. 19, No. 2, 483--515 (1984; Zbl 0549.14012)]. So the Severi inequality became the Severi conjecture. In the 80's Severi conjecture was the main motivation of \textit{G. Xiao's} work on surfaces fibred over a curve [Math. Ann. 276, No. 3, 449--466 (1987; Zbl 0596.14028)], where he proved the slope inequality. Xiao's paper does in fact prove Severi inequality for surfaces having an irrational pencil. In the 90's \textit{K. Konno} [Nagoya Math. J. 143, 1--11 (1996; Zbl 0872.14033] proved the conjecture in the special case of even surfaces, that is of surfaces whose canonical class is \(2-\) divisible in the Picard group. Finally, at the end of the 90's, \textit{M. Manetti} [Math. Nachr. 261/262, 105--122 (2003; Zbl 1044.14017)] could prove the inequality for the large class of surfaces of general type whose canonical bundle is ample. Pardini's idea is surprisingly easy: starting by a surface \(S\) with maximal Albanese dimension she constructs a sequence of genus \(g_d\) fibrations \(f_d:Y_d \rightarrow {\mathbb P}^1\) such that \[ \lim_{d \rightarrow \infty} g_d =+\infty \] and whose slope \(\lambda(f_d)\) verifies \[ \lim_{d \rightarrow \infty} \lambda(f_d) = K^2_S/\chi. \] Then, pushing to the limit Xiao's slope inequality \(\lambda (f_d) \geq 4(1-1/g_d)\), one gets exactly the desired inequality \(K^2_S/\chi \geq 4\).