No source-free exchange-correlation magnetic fields in non-collinear spin DFT

Abstract

We show that the Maxwellian equation $\nabla\cdot\mathbf{B}_{\mathrm{xc}}=0$, where $\mathbf{B}_{\mathrm{xc}}$ is the exchange-correlation (xc) magnetic field, is not an exact condition of non-collinear spin density functional theory: it violates global spin-rotation symmetry, the operation which turns the electron spins. How much does this matter in practice? To find out we impose the condition anyway, in the best way we can: we modify any parent xc functional so that the Kohn-Sham procedure yields a divergence-free $\mathbf{B}_{\mathrm{xc}}$ that is a functional derivative, stays exact for the homogeneous gas, and exerts the local torques a locally collinear functional cannot. Any non-physical result from the calculation must be due to the spurious condition, not the implementation. On Mn$_2$ with LSDA as parent, the source-free (SoF) construction and locally collinear (LoC) LSDA both give too short a bond and too large a bond energy, but differ on the magnetic exchange coupling and on the state itself: SoF finds the experimentally observed $^1Σ^+_g$ antiferromagnet, with a coupling of the right sign and roughly the right size, while LoC finds a high-spin $^{11}Π_u$ ferromagnet and the wrong sign. Despite this seeming success, breaking the symmetry has serious consequences: a rigid spin rotation of the magnetization changes $E_{\mathrm{xc}}$ by about 1 eV (it should not change at all), and the magnetization induced by a weak uniform field points perpendicular to the field (it should be antiparallel to it).