蒙特卡罗辐射转移方法在中、高光学深度的失效
The Failure of Monte Carlo Radiative Transfer at Medium to High Optical Depths
摘要:
光子通过吸收和/或散射介质传输的计算机模拟是天体物理学中的一个重要研究工具。几乎所有对三维几何形状进行这种模拟的软件代码都采用蒙特卡罗辐射转移方法,包括各种形式的偏置以加速计算。由于蒙特卡罗技术的概率性质,输出本身就有噪声,但通常假设平均值提供物理正确的结果。我们表明,这种假设并不总是合理的。具体来说,我们研究的辐射强度穿透一个无限的,均匀的平板材料,吸收和散射的辐射具有相同的概率。基本的蒙特卡罗辐射转移方法,没有任何偏置机制,开始崩溃的横向光学深度以上~20,因为如此少的模拟光子包到达另一侧的板。当包括诸如吸收/散射分裂和路径长度拉伸的偏置技术时,模拟的光子包确实到达板的另一侧,但是偏置权重不一定加起来就是正确的解。虽然噪声水平似乎是可以接受的,但平均值有时会严重低估正确的解决方案。检测这些异常需要明智地应用统计测试,类似于核粒子输运领域中使用的那些,可能与连续使用更大数量的光子包的收敛测试相结合。在任何情况下,对于横向光学深度超过~75,我们的研究中使用的蒙特卡罗方法无法解决一维板问题,这意味着需要近似,如修改后的随机游走。
英文摘要:
Computer simulations of photon transport through an absorbing and/or scattering medium form an important research tool in astrophysics. Nearly all software codes performing such simulations for three-dimensional geometries employ the Monte Carlo radiative transfer method, including various forms of biasing to accelerate the calculations. Because of the probabilistic nature of the Monte Carlo technique, the outputs are inherently noisy, but it is often assumed that the average values provide the physically correct result. We show that this assumption is not always justified. Specifically, we study the intensity of radiation penetrating an infinite, uniform slab of material that absorbs and scatters the radiation with equal probability. The basic Monte Carlo radiative transfer method, without any biasing mechanisms, starts to break down for transverse optical depths above ~20 because so few of the simulated photon packets reach the other side of the slab. When including biasing techniques such as absorption/scattering splitting and path length stretching, the simulated photon packets do reach the other side of the slab but the biased weights do not necessarily add up to the correct solution. While the noise levels seem to be acceptable, the average values sometimes severely underestimate the correct solution. Detecting these anomalies requires the judicious application of statistical tests, similar to those used in the field of nuclear particle transport, possibly in combination with convergence tests employing consecutively larger numbers of photon packets. In any case, for transverse optical depths above ~75 the Monte Carlo methods used in our study fail to solve the one-dimensional slab problem, implying the need for approximations such as a modified random walk.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。