The difficulty of getting high escape fractions of ionizing photons from high-redshift galaxies: a view from the FIRE cosmological simulations

We present a series of high-resolution (20-2000 M-circle dot, 0.1-4 pc) cosmological zoom-in simulations at z greater than or similar to 6 from the Feedback In Realistic Environment (FIRE) project. These simulations cover halo masses 10(9)-10(11) M-circle dot and rest-frame ultraviolet magnitude M-UV = -9 to -19. These simulations include explicit models of the multi-phase ISM, star formation, and stellar feedback, which produce reasonable galaxy properties at z = 0-6. We post-process the snapshots with a radiative transfer code to evaluate the escape fraction (f(esc)) of hydrogen ionizing photons. We find that the instantaneous f(esc) has large time variability (0.01-20 per cent), while the time-averaged f(esc) over long time-scales generally remains less than or similar to 5 per cent, considerably lower than the estimate in many reionization models. We find no strong dependence of f(esc) on galaxy mass or redshift. In our simulations, the intrinsic ionizing photon budgets are dominated by stellar populations younger than 3 Myr, which tend to be buried in dense birth clouds. The escaping photons mostly come from populations between 3 and 10 Myr, whose birth clouds have been largely cleared by stellar feedback. However, these populations only contribute a small fraction of intrinsic ionizing photon budgets according to standard stellar population models. We show that f(esc) can be boosted to high values, if stellar populations older than 3 Myr produce more ionizing photons than standard stellar population models (as motivated by, e.g. models including binaries). By contrast, runaway stars with velocities suggested by observations can enhance f(esc) by only a small fraction. We show that 'sub-grid' star formation models, which do not explicitly resolve star formation in dense clouds with n >> 1 cm(-3), will dramatically overpredict f(esc).