1.School of Nuclear Science and Engineering, Shanghai Jiao Tong University, Shanghai 200240, China 2.State Power Investment Central Research Institute Nuclear Power Software Development Center, Beijing 100029, China
Fund Project:Project supported by the National Science and Technology Major Project of the Ministry of Science and Technology of China (Grant No. 2017ZX06002002)
Received Date:26 April 2019
Accepted Date:06 August 2019
Available Online:01 October 2019
Published Online:20 October 2019
Abstract:Traditionally, the numerical performance of the lattice Boltzmann method is mainly determined by the moment degree of a discrete equilibrium distribution. The equilibrium distribution positivity is merely considered as an ancillary property which is used to constrict the numerical configuration. With the newly-developed partial Gaussian-Hermite quadrature scheme, the positivity of equilibrium distribution is validated as an independent property like moment degree which can be adjusted by discrete velocities. Researchers speculated that the positivity should also be significant for the numerical performance by the lattice Boltzmann method and can be used to improve the performance. Comparing with the classical improvement through moment degree, the positivity approach will not bring additional computation. However, due to the lack of boundary treatment, the speculation has not been validated by detailed numerical simulations. In this paper, through employing a periodic case, the Taylor-Green vortex, to avoid the boundary issue, we in depth analyze the numerical effect of the model positivity, including the numerical accuracy in the model positive range, the influence of positivity on the numerical performance, and the significance comparison between positivity and moment degree. The results show that for a given model, the numerical accuracy is not consistent in the whole positive range. As the configuration is close to the border of positive range, the accuracy will degrade though it is still acceptable. The numerical performance of a model depends on both moment degree and positivity. The role that the moment degree plays lies mainly in the qualification of a model on Galilean invariance. Once a model fulfills the Galilean invariance, its numerical performance is solely dependent on the positivity. Hence, the improvement approach through modifying the model positivity is a viable solution, and a Galilean invariant model with wider positive range does possess a better numerical performance regardless of its moment degree. Furthermore, based on the numerical results in this paper, all DnHm models with high moment degree are better than the classical D2Q9 model. Of the above models, the D2H3-2 model has the best performance and deserves to be further studied Keywords:lattice Boltzmann/ discrete model/ moment degree/ positivity
表2格子Boltzmann离散模型. 为方便展示, 表格罗列的是用于对应构造高阶模型的一维模型参数. 实际计算中需根据(9)式和(10)式对表格中的参数进行张量构造. 另外需注意的是表格中罗列的是网格常数c, 其与格子声速$ c_{\rm s} $存在如下换算关系$ c = 1/\sqrt{2}c_{\rm s} $ Table2.Discrete model description of lattice Boltzmann method. For the seeking of space saving, the parameters illustrated in this table are of the corresponding unidimensional models. In numerical implementations, they require tensor product illuminated in Eq. (9) and Eq. (10). It should be noted that the table lists the lattice constant c instead of the lattice sonic speed $ c_{\rm s} $, which can be expressed as $ c = 1/\sqrt{2}c_{\rm s} $.
$ {\kappa} = \frac{{\displaystyle\sum {\left| f \right|} }}{{\displaystyle\sum f }}=\frac{{\displaystyle\sum {\left| f \right|} }}{\rho }. $
当分布中存在负值时, $ \kappa $值将大于1, 而不存在负值时则$ \kappa $值恒为1. $ \kappa $偏离1的幅值可以看作分布正值性破坏的严重程度. 该指标通过对$ \rho $归一, 消除了Taylor-Green涡流场内不同密度对正值性的影响. 图2给出了设计计算工况下, 各模型初始分布的$ \kappa $值情况. 图中额外标注了对应工况下模型正值区域值$ u_{\rm pos} $与设计$ u_{L{\rm B} ,0} $的差值, 记作$ \Delta u $, 图 2 各格子Boltzmann模型在Taylor-Green涡设计计算工况下初始时刻计算流域$ [0,2{\text{π}}]\times[0,2{\text{π}}] $内$ \kappa $值色阶图 图中列对应格子Boltzmann模型, 行则对应表3中计算工况; 每个色阶图上方的数值$ \Delta u $值, 即正值区域值$ u_{\rm pos} $设计计算工况$ u_{L{\rm B} ,0} $差值;为方便观察正值性随计算工况遭到破坏的变化过程, 图中对$ \kappa $值做了截断处理, 仅区别$ [1.0,2.0] $范围变化 Figure2. The initial $ \kappa $ filled contours of designed Taylor-Green vortex simulations for each lattice Boltzmann model in the fluid domain $ [0,2{\text{π}}]\times[0,2{\text{π}}] $. Each column is a simulation set of a lattice Boltzmann model with different designed cases. Each row is a set of a simulation case with different lattice Boltzmann models. The values in the contour panels are the values of $ \Delta u $, i.e. the difference between the $ u_{\rm pos} $ and the designed $ u_{L{\rm B} ,0} $. For the sake of identifying the positivity violation process of a lattice Boltzmann model, the filled contours truncate the range of $ \kappa $, limiting into $ [1.0,2.0] $.
鉴于速度平方和$ U^2 $在流域内成周期性分布, 本文仅截取周期区域$ [0.3775,0.6225] {\text{π}} \times[0.3775,0.6225] {\text{π}} $内图像(见图3). 误差演化分析则关注模型计算结果与理论解误差在$ t = 3.4657359 \;\rm{s} $时间内随时间的演变(图4), 其计算公式为 图 3 Taylor-Green涡半衰时刻$ t = 3.4657359\; \rm{s} $, $ [0.3775,0.6225]{\text{π}} \times[0.3775,0.6225]{\text{π}} $流域内, 不同计算工况下理论解(Theoretical) 与各格子Boltzmann模型的速度平方和$ U^2 $等高线图; 图中列对应格子Boltzmann模型, 行则对应表3中计算工况 Figure3. The $ U^2 $ contour lines of Taylor-Green vortex in fluid domain $ [0.3775,0.6225]{\text{π}} \times[0.3775,0.6225]{\text{π}} $ at half-value decay time $ t = 3.4657359 \; \rm{s} $, including the theoretical solution (denoted as “Theoretical”) and the numerical results of lattice Boltzmann models. Each column is a set of simulations under the designed cases. Each row is a set of simulations under selected lattice Boltzmann models including the theoretical solution.
图 4 各Taylor-Green涡设计计算工况下, 各格子Boltzmann模型计算误差随时间演变, 图中标号a, b, c, d分别对应表3各设计计算工况; 为方便表示, 图中横坐标单位为$ 0.173286795\; \rm{s} $ Figure4. The error evolution of each simulation. The panel a, b, c, d renders each designed case in Table 3 respectively. For the sake of rendering, the unit of the time axis is scaled as 0.173286795 s.
式中$ u_a $和$ v_a $为理论结果. 从计算结果来看, 对于每个DnHm模型, 其数值精度随着$ u_{L{\rm B} ,0} $增大而变坏, 无论是计算结果等高线与理论解的相似程度(图3), 还是误差演化(图4)都体现了该趋势. 但与模型正值性一受到破坏计算即发散的D2H2模型不同的是, 高阶DnHm模型对于模型正值性破坏并未体现出间断式的数值表现变化. 对于单个计算工况, DnHm系列模型除不满足Galilean不变性的D2H2模型外, 计算精度随$ u_{\rm pos} $减小而变坏. 而D2H2模型精度则差于所有高阶模型, 包括正值区域更小的D2H4模型. 对比D2H2模型与其他高阶模型的数值表现, 模型的Galilean不变性对于模型的计算稳定性具有显著影响. 从这一角度而言, 矩精度对模型数值表现影响主要体现在模型是否满足Galilean不变性上. 对于满足Galilean不变性的模型, 矩精度影响不再显著, 模型数值表现主要受正值性影响. 分析模型正值区域内计算表现, 即$ \Delta u $大于0的数值模拟, 所有计算结果等高线与理论解均保持较好的符合, 而对比误差(图5), 计算结果未明显表现出模型正值区域与精度之间跨工况相关性. 具体而言, 在模型正值区域内, 具有较宽正值区域的模型并不对计算恒定保持优势精度, 以D2H3-2模型工况b为例, 其计算差于所有其他3阶及3阶以上DnHm模型工况a计算结果. 进一步而言, 在模型正值区域内, 其计算精度并不是恒定的, 计算工况越靠近正值区域上限, 模型精度越低. 因此模型正值区域优势主要体现在同一计算工况对比. 图 5 Taylor-Green涡设计计算工况下, $ \Delta u $值为正值的各格子Boltzmann模型数值模拟误差对比. 图中实线, 虚线和虚点线分别为工况a, b, c (见表3)结果, 离散模型则用颜色和符号标注. 为方便表示, 图中横坐标单位为0.173286795 s Figure5. The error comparison of Taylor-Green vortex simulations with postive value of $ \Delta u $. The designed configurations of Taylor-Green vortex are denoted with line styles, in which the solid, dashed and dash-dotted line indicates case a, b and c in Table 3 respectively, meanwhile the lattice Boltzmann models are labeled with line colors and markers. For the sake of rendering, the unit of the time axis is scaled as 0.173286795 s.
在数值计算中, 独立的精度指标是非常实用的指示参量, 即对于任意计算工况, 只需保证该精度指标一致, 则计算精度一致. 鉴于正值区域的精度优势主要体现在同工况下模型之间的对比, 无法表征数值模拟的确切精度, 本文尝试探索其他参量作为精度指标的可能性. 在Taylor-Green涡算例中, 文章设计了$ \Delta u $值和$ \kappa $值来表征模型在对应计算工况下的正值性情况. 本文探索这两指标作为衡量计算精度的可能性. 文章首先检验$ \Delta u $值, 以D2H4模型工况a, b, c与D2H5模型工况b, c, d为例, 两两分别具有近似的$ \Delta u $值. 若格子Boltzmann方法的数值表现与$ \Delta u $值强相关, 则D2H4模型与D2H5模型对应算例精度将呈现明显的相关性. 从图6 (a)来看, 对比D2H3-2模型随工况变化而导致精度变化程度, 对应D2H4模型与D2H5模型结果之间的精度差异幅度依旧明显, 并不足以体现格子Boltzmann方法计算精度与$ \Delta u $值相关性. 另外取$ \Delta u $值在$ (-0.30\pm0.6) $附近计算结果 (图6 (b)), 同样对比D2H3-2模型随工况变化而导致精度变化程度, 选取的计算精度并未收敛于足够识别$ \Delta u $值影响的值域附近. 从上述分析可以看出, $ \Delta u $值并不具备衡量格子Boltzmann方法数值精度的能力. 对于$ \kappa $值, 本文选取了模型正值性已遭受破坏但初始$ \kappa $值尚未体现显著偏离的计算结果 (图7). 类似地, 对比D2H3-2模型随工况变化而导致精度变化程度, 误差对比显示, 选取算例并未显示显著的收敛性, 因此$ \kappa $值也不具备衡量格子Boltzmann方法数值表现的能力. 图 6 Taylor-Green涡数值模拟误差与$ \Delta u $值相关性分析 (a)对比了$ \Delta u $接近的D2H4模型a, b, c工况与D2H5模型b, c, d工况计算误差; (b)对比了$ \Delta u $在$ -0.30 $附近所有数值模拟计算误差. 所有图中均保留了D2H3-2模型在a, b, c, d工况下的计算误差作为参照. 曲线上标注值为该数值模拟的$ \Delta u $值. 图中a, b, c, d工况分别用实线, 虚线, 虚点线及点线标注, 而离散模型则用颜色和符号标注. 为方便表示, 图中横坐标单位为0.173286795 s Figure6. The numerical performance of Taylor-Green vortex simulations vs the value of $ \Delta u $. Panel a plots the numerical errors of model D2H4 under case a, b, c and model D2H5 under case b, c, d, which possess close value of $ \Delta u $ respectively. Panel b plots the numerical errors of simulations with a value of $ \Delta u $ around $ -0.30 $. The numbers labeled on the curves are their values of $ \Delta u $. The simulation configurations are denoted with line style, in which solid, dashed, dash-dotted and dotted line indicates case a, b, c and d respectively, meanwhile the lattice Boltzmann models are labeled with line colors and markers. In all panels, the results of model D2H3-2 with four designed configurations are plotted for a reference. The designed configurations of Taylor-Green vortex are denoted with line styles, in which the solid, dashed, dash-dotted and dotted line indicates case a, b, c and d in Table 3 respectively, meanwhile the lattice Boltzmann models are labeled with line colors and markers. For the sake of rendering, the unit of the time axis is scaled as 0.173286795 s.
图 7 Taylor-Green涡数值模拟误差与初始$ \kappa $值相关性分析. 图中对比了在$ \Delta u $值为负情况下, 初始$ \kappa $值偏离$ 1 $幅值可忽略的算例计算误差, 包括模型D2H3-1工况c, 模型D2H3-2工况d, 模型D2H4工况b, 模型D2H5工况c, d和模型D2H6工况b, c. 图中均保留了D2H3-2模型在a, b, c, d工况下的计算误差作为参照. 曲线上标注值为该数值模拟的$ \Delta u $值. 图中a, b, c, d工况分别用实线, 虚线, 虚点线及点线标注, 而离散模型则用颜色和符号标注. 为方便表示, 图中横坐标单位为0.173286795 s Figure7. The numerical performance of Taylor-Green vortex simulations vs initial $ \kappa $. The error evolutions of numerical simulations with negative values of $ \Delta u $ but negligible departure of initial $ \kappa $ from $ 1 $ are plotted, including model D2H3-1 under case c, model D2H3-2 under case d, model D2H4 under case b, model D2H5 under case c,d and model D2H6 under case b, c. The results of model D2H3-2 with four designed configurations are also plotted for a reference. The designed configurations of Taylor-Green vortex are denoted with line styles, in which the solid, dashed and dash-dotted line indicates case a, b and c in Table 3 respectively, meanwhile the lattice Boltzmann models are labeled with line colors and markers. For the sake of rendering, the unit of the time axis is scaled as 0.173286795 s.