The $H\to \tau \tau$ decays form a promising channel for tests of the $CP$ invariance of Higgs boson couplings. A previous study has shown the viability of deep learning techniques for such a measurement. In this paper, the study is extended to consider potential sources of uncertainty. Effects due to the limited experimental resolution are discussed. Furthermore, systematics due to $\tau$ decay modeling for complex cascade decays to ${\tau }^{\pm }\to a_1^{\pm }{\nu }_{\tau }\to {\rho }^0{\pi }^{\pm }{\nu }_{\tau }\to 3{\pi }^{\pm }{\nu }_{\tau }$ are also addressed. Various parametrizations using low-energy collision data are considered.

• Received 17 July 2017

DOI:https://doi.org/10.1103/PhysRevD.96.073002