Singularities with the highest Mather minimal log discrepancy (Q2435109)

Minimal log discrepancies are a natural and subtle measure of the singularities of pairs \((X,B)\) that appear in the context of the minimal model program. (Recall that if \((X,B)\) is a pair, then \(X\) is a normal variety and \(B\) is an effective \(\mathbb R\)-divisor such that \(K_X+B\) is \(\mathbb R\)-Cartier.) A famous conjecture of Shokurov states that if \(x\) is a closed point on a pair \((X,B)\) then the minimal log discrepancy satisfies the inequality \(\mathrm{mld}(x;X,B)\leq \dim X\) and equality holds if and only if \((X,x)\) is non-singular and \(B=0\) near \(x\). Unluckily, minimal log discrepancies are hard to study and the above conjecture is widely open. Mather minimal log discrepancies \(\widehat {\mathrm{mld}}(x;X,\mathcal J _XB)\) are defined in terms of the Jacobian ideal \(\mathcal J_X\) (the hypothesis that \(K_X+B\) is \(\mathbb R\)-Cartier no longer necessary), and provide an alternative and more tractable point of view. In this paper, the authors prove that Mather minimal log discrepancies satisfy the inequality \(\widehat {\mathrm{mld}}(x;X,\mathcal J _XB)\leq \dim X\) and equality holds if and only if \((X,x)\) is non-singular and \(B=0\) near \(x\). The authors also give a classification of singularities with \(\dim X -1 \leq \widehat {\mathrm{mld}}(x;X,\mathcal J _XB)\). Since the Mather minimal log discrepancy coincides with the usual minimal log discrepancy when \(X\) is normal and locally complete intersection, these results imply Shokurov's conjecture for normal l.c.i. varieties.