Hostname: page-component-7b7567c8c-wxvwp Total loading time: 0 Render date: 2026-11-16T07:09:15.752Z Has data issue: false hasContentIssue false

Hyperbolicity, shadowing directions and sensitivity analysis of a turbulent three-dimensional flow

Published online by Cambridge University Press:  28 January 2019

Angxiu Ni*
Affiliation:
Department of Mathematics, University of California, Berkeley, Berkeley, CA 94720, USA
*
†Email address for correspondence: niangxiu@gmail.com

Abstract

This paper uses compressible flow simulation to analyse the hyperbolicity, shadowing directions and sensitivities of a weakly turbulent three-dimensional cylinder flow at Reynolds number 525 and Mach number 0.1. By computing the first 40 covariant Lyapunov vectors (CLVs), we find that unstable CLVs are active in the near-wake region, whereas stable CLVs are active in the far-wake region. This phenomenon is related to hyperbolicity since it shows that CLVs point to different directions; it also suggests that for open flows there is a large fraction of CLVs that are stable. However, due to the extra neutral CLV and the occasional tangencies between CLVs, our system is not uniform hyperbolic. By the non-intrusive least-squares shadowing (NILSS) algorithm, we compute shadowing directions and sensitivities of long-time-averaged objectives. Our results suggest that shadowing methods may be valid for general chaotic fluid problems.

Information

Type
JFM Papers
Copyright
© 2019 Cambridge University Press 
Figure 0

Figure 1.Geometry used in the simulation of a 3D flow past a cylinder. The spanwise extent of the computational domain is $Z=2D$. The positive direction of the cylinder rotational speed $\unicode[STIX]{x1D714}$ is anticlockwise.

Figure 1

Figure 2.(a) A 2D slice of the finer mesh over the entire computational domain. (b) A zoom around the cylinder. This is a block-structured mesh with $7.5\times 10^{5}$ hexahedra. The spanwise direction has 48 cells.

Figure 2

Figure 3.Some 2D slices of the flow field on the finer mesh. (a) Vertical cross-section, plotted by the magnitude of velocity. (b) Horizontal cross-section, plotted by the spanwise velocity; this shows that the flow is 3D. All velocities are normalized by the free-stream velocity $U_{0}=33.0$.

Figure 3

Figure 4.Flow field on the finer mesh averaged over a time span of $168t_{0}$, plotted by the $\unicode[STIX]{x1D70C}\boldsymbol{U}$ component, normalized by $\unicode[STIX]{x1D70C}_{0}U_{0}$.

Figure 4

Table 1.Comparison with previous literature for the Strouhal number $S_{t}$ and the averaged drag coefficient $C_{D}$.

Figure 5

Figure 5.Lyapunov exponents (LE) normalized by $t_{0}^{-1}$. Left: convergence history of the first 40 different LEs. Right: confidence intervals of LEs. (a) Results on the coarser mesh. The largest LE is $0.20t_{0}^{-1}$, meaning in one flow-through time $t_{0}$, the norm of the first CLV becomes $\text{e}^{0.20}=1.22$ times larger. (b) Results on the finer mesh. The largest LE is $0.21t_{0}^{-1}$.

Figure 6

Figure 6.Histogram of angles between CLVs on segments 350–449, with the entire trajectory of length $T=211t_{0}$ partitioned into $K=800$ segments. The total area of the histogram is normalized to one. (a) Coarser mesh, all CLVs, smallest angle is $5.8^{\circ }$. (b) Coarser mesh, CLV indices more than five apart, smallest angle is $29.2^{\circ }$. (c) Finer mesh, all CLVs, smallest angle is $16.4^{\circ }$. (d) Finer mesh, CLV indices more than five apart, smallest angle is $40.8^{\circ }$.

Figure 7

Figure 7.Minimal and averaged angles between all pairs of CLVs. The minimum and average are taken on segments 350–449, with the entire trajectory of length $T=211t_{0}$ partitioned into $K=800$ segments. (a) Coarser mesh, minimal angles between all pairs of CLVs. (b) Coarser mesh, averaged angles between all pairs of CLVs. (c) Finer mesh, minimal angles between all pairs of CLVs. (d) Finer mesh, averaged angles between all pairs of CLVs.

Figure 8

Figure 8.CLVs computed on the coarser mesh on a trajectory of time length $T=211t_{0}$: (a) 1st CLV; (b) 40th CLV; and (c) primal flow field. Left: snapshots at $T/2$. Right: averaged over time span [$95t_{0},116t_{0}$]. Plotted by the magnitude of $\unicode[STIX]{x1D70C}\boldsymbol{U}$ component, and normalized such that the largest value is 1.00. The primal flow at the bottom is normalized by $\unicode[STIX]{x1D70C}_{0}U_{0}$.

Figure 9

Figure 9.CLV computed on the finer mesh, using same settings as the coarser mesh: (a) 1st CLV; (b) 5th CLV; (c) 17th CLV; (d) 40th CLV; and (e) primal flow field.

Figure 10

Figure 10.Shadowing directions. The base trajectory has parameter $s$, the shadowing trajectory has parameter $s+\unicode[STIX]{x1D6FF}s$. The first-order approximation of the perpendicular distance is $\unicode[STIX]{x1D6FF}\boldsymbol{u}^{\bot }\approx \boldsymbol{v}^{\infty \bot }\unicode[STIX]{x1D6FF}s$, where $\boldsymbol{v}^{\infty \bot }$ is the shadowing direction.

Figure 11

Figure 11.Norms of shadowing directions on the finer mesh, measured from $66t_{0}$ to $93t_{0}$, computed on a trajectory of length $T=158t_{0}$: (a) with $U$ as parameter; and (b) with $\unicode[STIX]{x1D714}$ as parameter.

Figure 12

Figure 12.Magnitude of the $\unicode[STIX]{x1D70C}\boldsymbol{U}$ component of the shadowing direction $\boldsymbol{v}^{\bot }=\text{d}\boldsymbol{u}^{\bot }/\text{d}s$ on the finer mesh, which is computed by FD-NILSS on a trajectory of time length $T=158t_{0}$: (a) $U$ as the parameter, normalized by $\unicode[STIX]{x1D70C}_{0}U_{0}/U_{0}=\unicode[STIX]{x1D70C}_{0}$; (b) $\unicode[STIX]{x1D714}$ as the parameter, normalized by $\unicode[STIX]{x1D70C}_{0}U_{0}/\unicode[STIX]{x1D714}_{0}$; and (c) primal flow field, normalized by $\unicode[STIX]{x1D70C}_{0}U_{0}$. Left: snapshots at $0.5T$; right: averaged from $66t_{0}$ to $93t_{0}$. Notice that the left of (a) and (b) are coloured by log scale.

Figure 13

Figure 13.History plots of sensitivities computed by FD-NILSS: (a) sensitivity of the drag with respect to $U$, and (b) sensitivity of the lift with respect to $\unicode[STIX]{x1D714}$.

Figure 14

Figure 14.Sensitivities for different choices of parameters and objectives: (a) drag versus $U$, and (b) lift versus $\unicode[STIX]{x1D714}$. Vertical bars indicate 95 % confidence intervals of the averaged objectives. The straight lines though the middle data points indicate the sensitivities computed by FD-NILSS.

Supplementary material: File

Ni supplementary material

Ni supplementary material 1

Download Ni supplementary material(File)
File 5.1 MB