Computational speeds are compared for modern implementations of three parabolic equation approximations. The split-step c0-insensitive model is 3.7 and 5.5 times faster than the finite-difference model OWWE and finite-element model RAM, respectively. Calculations are made between a source and receiver separated horizontally by 1000[Formula: see text]km at 600[Formula: see text]m depth near the minimum of sound speed in the deep ocean. Sound speed varies with depth but not range. The impulse response is computed by applying an inverse Fourier transform to equispaced discrete frequencies between 50[Formula: see text]Hz and 100[Formula: see text]Hz. At convergence, all implementations yield multipath travel times and transmission losses within 7[Formula: see text]ms and 2.4[Formula: see text]dB of an exact solution computed with normal modes via KRAKEN.

中文翻译：

---

三个抛物线近似的计算速度

比较了三个抛物线方程近似的现代实现的计算速度。分步 c0 不敏感模型分别比有限差分模型 OWWE 和有限元模型 RAM 快 3.7 倍和 5.5 倍。计算是在水平间隔 1000[公式：见文本]km 处的源和接收器之间进行的，深度为 600[公式：见文本]m，深度接近深海中声速的最小值。声速随深度而变化，但不随范围变化。通过对 50[公式：参见文本]Hz 和 100[公式：参见文本]Hz 之间的等距离散频率应用傅里叶逆变换来计算脉冲响应。在收敛时，所有实现在 7[公式：见文本]ms 和 2.4[公式：