Definition 11.4.1.70139B6ABDC2 A Hermitian form $\gamma $ is Gauduchon, or a Gauduchon metric, if
It is balanced if $\mathrm d(\gamma ^{n-1})=0$, and pluriclosed if $\mathrm{dd}^{\mathrm{c}}\gamma =0$.
4ECC3E68F7D54ECC3E68F7D5
By proposition 11.1.8, the first variation of the mass at $u=0$ vanishes exactly when $\mathrm{dd}^{\mathrm{c}}(\omega ^{n-1})=0$. This condition can be achieved in every conformal class: finding a conformal factor amounts to finding a positive generator of the kernel of the adjoint of a scalar elliptic operator.
Definition 11.4.1.70139B6ABDC2 A Hermitian form $\gamma $ is Gauduchon, or a Gauduchon metric, if
It is balanced if $\mathrm d(\gamma ^{n-1})=0$, and pluriclosed if $\mathrm{dd}^{\mathrm{c}}\gamma =0$.
Theorem 11.4.2 (Gauduchon's theorem).3978AEB15BD7 Every Hermitian form $\eta $ has a conformal Gauduchon representative. More precisely, there is a smooth strictly positive function $\rho $, unique up to multiplication by a positive constant, such that
The associated Gauduchon form is $\gamma =\rho ^{1/(n-1)}\eta $.
Proof. For a smooth real function $v$ on $X$, define
This real second-order operator has positive definite principal symbol and no zeroth-order term. By the maximum principle, its kernel consists of the constants. Its index is zero, since the linear path joining it to a Laplace operator consists of elliptic operators. Fredholm theory and elliptic regularity therefore show that the formal adjoint $L^*$ of $L$ with respect to $\eta ^ n$ has a one-dimensional kernel, generated by a nonzero smooth real function $\rho $.
We first prove that $\int _ X\rho \eta ^ n\ne 0$. Otherwise the constant function $1$ would be orthogonal to the kernel of $L^*$, and the Fredholm alternative would solve $Lv=1$, which is impossible at a maximum point of $v$. We may therefore rescale so that $\int _ X\rho \eta ^ n=1$.
Next we show that $\rho \geq 0$. Suppose instead that $\rho $ is negative somewhere. Choose a nonzero smooth function $h\geq 0$ supported where $\rho <0$, and put $c=\int _ Xh\rho \eta ^ n<0$. The function $h-c$ is strictly positive and orthogonal to the kernel of $L^*$, so the Fredholm alternative would solve $Lv=h-c>0$, which is again impossible at a maximum point of $v$.
In real local coordinates $(x_1,\ldots ,x_{2n})$, write $L^*=A^{ij}\partial _ i\partial _ j+B^ i\partial _ i+C$, where $\partial _ i=\partial /\partial x_ i$, the coefficients are smooth, and repeated indices are summed from $1$ to $2n$. The leading symbol $(A^{ij})$ is positive definite. Since $\rho \geq 0$,
and the operator $L^*-\max (C,0)$ has nonpositive zeroth-order term. By the strong minimum principle, a zero of $\rho $ would force $\rho $ to vanish identically on the connected manifold $X$. Hence $\rho >0$.
Finally, integration by parts gives
Thus the equation $L^*\rho =0$ is precisely $\mathrm{dd}^{\mathrm{c}}(\rho \eta ^{n-1})=0$, and the one-dimensional adjoint kernel gives uniqueness up to positive scaling. Since $\gamma ^{n-1}=\rho \eta ^{n-1}$, the form $\gamma $ is Gauduchon. $\square$
A Kähler form, that is, a closed Hermitian form, is both balanced and pluriclosed, and every balanced form is Gauduchon. On surfaces, balanced forms are Kähler, and the Gauduchon and pluriclosed conditions coincide.
Comments
Corrections, questions, and mathematical discussion are welcome. Sign in with GitHub to comment. Comments are public.