To investigate the potential abundance and impact of nuclear black holes (BHs) during reionization, we generate a neural network that estimates their masses and accretion rates by training it on 23 properties of galaxies harbouring them at $z=6$ in the cosmological hydrodynamical simulation Massive-Black II. We then populate all galaxies in the simulation from $z=18$ to $z=5$ with BHs from this network. As the network allows to robustly extrapolate to BH masses below those of the BH seeds, we predict a population of faint BHs with a turnover-free luminosity function, while retaining the bright (and observed) BHs, and together they predict a Universe in which intergalactic hydrogen is $15\%$ ionized at $z=6$ for a clumping factor of 5. Faint BHs may play a stronger role in H reionization without violating any observational constraints. This is expected to have an impact also on pre-heating and -ionization, which is relevant to observations of the 21 cm line from neutral H. We also find that BHs grow more efficiently at higher $z$, but mainly follow a redshift-independent galaxy-BH relation. We provide a power law parametrisation of the hydrogen ionizing emissivity of BHs.

中文翻译：

---

在再电离过程中模拟微弱的黑洞

为了研究再电离过程中核黑洞 (BH) 的潜在丰度和影响，我们生成了一个神经网络，通过在宇宙流体动力学模拟中以 $z=6$ 的星系的 23 个属性对其进行训练，来估计它们的质量和吸积率Massive-Black II。然后，我们使用来自该网络的 BH 填充模拟中从 $z=18$ 到 $z=5$ 的所有星系。由于网络允许稳健地外推到低于 BH 种子质量的 BH 质量，我们预测了一群具有无周转光度函数的微弱 BH，同时保留了明亮的（和观察到的）BH，并且它们一起预测了一个宇宙，其中星系间氢在 $z=6$ 处电离 $15\%$，聚集因子为 5。微弱的 BHs 可能在 H 再电离中发挥更大的作用，而不会违反任何观测限制。预计这也会对预热和电离产生影响，这与观察中性 H 的 21 厘米线有关。我们还发现 BH 在 $z$ 较高时更有效地生长，但主要遵循红移-独立的星系-BH 关系。我们提供了 BH 的氢电离发射率的幂律参数化。