1887
banner image
No data available.
Please log in to see this content.
You have no subscription access to this content.
No metrics data to plot.
The attempt to load metrics for this article has failed.
The attempt to plot a graph for these metrics has failed.
Characterizing chaotic dynamics from simulations of large strain behavior of a granular material under biaxial compression
Rent:
Rent this article for
USD
10.1063/1.4790833
/content/aip/journal/chaos/23/1/10.1063/1.4790833
http://aip.metastore.ingenta.com/content/aip/journal/chaos/23/1/10.1063/1.4790833

Figures

Image of FIG. 1.
FIG. 1.

The Canadian Lynx time series: chaotic or periodic? The upper panel shows the original data (open circles) from which various global nonlinear models have been built. The lower panel depicts the dynamical behavior of three such models. The horizontal axis in the upper panel is the observation year: one data point per year. In the lower three panels the horizontal axis are in units of model time steps: 10 time steps per year. Hence, the upper panel covers a period of 115 years, the lower panels cover a period of 300 years. The vertical scales are arbitrary (derived from the total quantity of lynx pelts harvested in a given year). The dynamics observed in the lower three panels are: chaotic, almost periodic and exactly periodic.

Image of FIG. 2.
FIG. 2.

Biaxial compression test. (Top) The observed time series of the bulk measurement of the stress ratio with respect to axial strain . The strain interval of interest is indicated by the bolder blue trace and covers the high strain post-peak regime (so-called critical state), where the material has failed and the response is in an approximately steady-state, exhibiting characteristic stick-slip (jamming-unjamming) dynamics. (Bottom) Although not used in the modeling it is informative to soil mechanicians to observe the volumetric strain response evolution with respect to axial strain. The sample initially contracts and then dilates before reaching a more steady state response in the high strain region from where models in a reconstructed phase space of time-delay stress ratio variables are built.

Image of FIG. 3.
FIG. 3.

False nearest neighbors. The upper two panels depict the proportion of false nearest neighbors (embedding lag 1) for the data depicted in Fig. 2 . For there are no false nearest neighbors, the proportion drops to almost 0 for . The lower panel is the proportion of local false nearest neighbors, a measure of the locally sufficient embedding dimension, for local dimension (again, embedding lag is 1 and the prediction horizon is 3). The calculation is repeated for various n = local neighborhood sizes and plateau onset at a local dimension of is evident. The local Lyapunov exponent spectrum indicated four genuine exponents with one or two positive.

Image of FIG. 4.
FIG. 4.

Linearly filtered noise. Comparison of nonlinear statistics computed with the Gaussian Kernel algorithm (correlation dimension, entropy and noise level), and higher order linear distribution statistics (skewness and kurtosis), for the original data in Fig. 2 and linear surrogates (monotonic nonlinear transformations of linearly filtered noise). As expected the linear statistics (lower right plots) show no difference. However, the nonlinear measures indicate that the linear model does not adequately described the dynamics (upper panels and lower left). For the nonlinear measures, the solid blue lines (no error bars) indicate the statistic values computed for the data (as a function of the embedding dimension—a parameter of the statistic). The tight error bars (red) are the mean andstandard deviation from 100 simulations, the larger error bars (green) are the full range (minimum to maximum). For the higher order linear statistics, a distribution of values is plotted, the value for the data are indicated as an asterisk.

Image of FIG. 5.
FIG. 5.

Stable node. The data displayed here are in the same format as Fig. 4 , except here the 100 simulations come from a model of the data driven by noise. In the absence of noise the model exhibits a stable node.

Image of FIG. 6.
FIG. 6.

Stable focus. The data displayed here are in the same format as Fig. 4 , except here the 100 simulations come from a model of the data driven by noise. In the absence of noise the model exhibits a stable focus.

Image of FIG. 7.
FIG. 7.

Transient chaos. The data displayed here are in the same format as Fig. 4 , except here the 100 simulations come from a model of the data driven by noise. In the absence of noise the model exhibits a chaotic transient (typically over a time scale longer than the observed data) and a stable fixed point.

Image of FIG. 8.
FIG. 8.

Motif frequency figure. For each of the three model dynamics we computed the motif super-family which occurred most frequently—shown here as a percentage. Motif family ADBCEF is the corresponding family for the data (the right-most batch of columns). In models exhibiting a stable focus, for example Motif ABDCEF occurred for all but one of the model simulations, that exception was ADBCEF. According to this measure, stable node and transient chaos models exhibited behavior most similar to the data. Inset: the structure of the six four node motifs. The motif-superfamily is a ranking by frequency of sub-graphs of order 4. The inset enumerates the six different motifs of order 4 which are possible. The horizontal axes of the upper panel is a ranking of these six classes (from most frequent to least).

Image of FIG. 9.
FIG. 9.

Sample trajectories. The top panel depicts the original data. The following four panels are iterated free run simulations, with dynamical noise for four different simulations from the model exhibiting transient chaos and the same motif super-family as the data.

Image of FIG. 10.
FIG. 10.

Sample trajectory. The top panel depicts a single noise free simulation exhibiting transient chaos dynamics. The lower panel is an embedding of the same data (color coded to depict different dynamical regimes). Note that the system appears to switch between two different dynamical behaviors before eventually collapsing to a fixed point. The time scale is the same as Fig. 9 .

Image of FIG. 11.
FIG. 11.

Sample trajectory. The left panel depicts a short section of the embedded data from Fig. 10 —the section is representative of the transient chaotic state and is the purely deterministic model output (no noise). On the right is the complex network constructed from this embedding.

Tables

Generic image for table
Table I.

DEM 2D biaxial compression test parameters and material properties.

Generic image for table
Table II.

Complex network statistics. For the original data and each of the three model classes, we computed observed values of the various complex network statistics: mean path-length (aka diameter), clustering, and assortativity. In each case, the mean value and standard deviation are reported. Also reported is the number of simulations (out of 100 attempts) which remained bounded. According to these measures both transient chaos and stable node dynamics are in agreement for two out of the three measures and differ marginally for the third.

Loading

Article metrics loading...

/content/aip/journal/chaos/23/1/10.1063/1.4790833
2013-02-08
2014-04-21
Loading

Full text loading...

This is a required field
Please enter a valid email address
752b84549af89a08dbdd7fdb8b9568b5 journal.articlezxybnytfddd
Scitation: Characterizing chaotic dynamics from simulations of large strain behavior of a granular material under biaxial compression
http://aip.metastore.ingenta.com/content/aip/journal/chaos/23/1/10.1063/1.4790833
10.1063/1.4790833
SEARCH_EXPAND_ITEM