return to top
source
√td(X) and the square root of a normalized graded class. The Mukai vector v(E) = ch(E)·√td(X) builds on this.
√td(X)
v(E) = ch(E)·√td(X)