An official website of the United States government
Inspired by the recent realization of a two-dimensional (2-D) chiral fluid as an active monolayer droplet moving atop a 3-D Stokesian fluid, we formulate mathematically its free-boundary dynamics. The surface droplet is described as a general 2-D linear, incompressible and isotropic fluid, having a viscous shear stress, an active chiral driving stress and a Hall stress allowed by the lack of time-reversal symmetry. The droplet interacts with itself through its driven internal mechanics and by driving flows in the underlying 3-D Stokes phase. We pose the dynamics as the solution to a singular integral–differential equation, over the droplet surface, using the mapping from surface stress to surface velocity for the 3-D Stokes equations. Specializing to the case of axisymmetric droplets, exact representations for the chiral surface flow are given in terms of solutions to a singular integral equation, solved using both analytical and numerical techniques. For a disc-shaped monolayer, we additionally employ a semi-analytical solution that hinges on an orthogonal basis of Bessel functions and allows for efficient computation of the monolayer velocity field, which ranges from a nearly solid-body rotation to a unidirectional edge current, depending on the subphase depth and the Saffman–Delbrück length. Except in the near-wall limit, these solutions have divergent surface shear stresses at droplet boundaries, a signature of systems with codimension-one domains embedded in a 3-D medium. We further investigate the effect of a Hall viscosity, which couples radial and transverse surface velocity components, on the dynamics of a closing cavity. Hall stresses are seen to drive inward radial motion, even in the absence of edge tension.
more »
« less
EikoNet: Solving the Eikonal Equation With Deep Neural Networks
The recent deep learning revolution has created enormous opportunities for accelerating compute capabilities in the context of physics-based simulations. In this article, we propose EikoNet, a deep learning approach to solving the Eikonal equation, which characterizes the first-arrival-time field in heterogeneous 3-D velocity structures. Our grid-free approach allows for rapid determination of the travel time between any two points within a continuous 3-D domain. These travel time solutions are allowed to violate the differential equation--which casts the problem as one of optimization--with the goal of finding network parameters that minimize the degree to which the equation is violated. In doing so, the method exploits the differentiability of neural networks to calculate the spatial gradients analytically, meaning that the network can be trained on its own without ever needing solutions from a finite-difference algorithm. EikoNet is rigorously tested on several velocity models and sampling methods to demonstrate robustness and versatility. Training and inference are highly parallelized, making the approach well-suited for GPUs. EikoNet has low memory overhead and further avoids the need for travel-time lookup tables. The developed approach has important applications to earthquake hypocenter inversion, ray multipathing, and tomographic modeling, as well as to other fields beyond seismology where ray tracing is essential.
more »
« less
- Award ID(s):
- 1822214
- PAR ID:
- 10226184
- Date Published:
- Journal Name:
- IEEE Transactions on Geoscience and Remote Sensing
- ISSN:
- 0196-2892
- Page Range / eLocation ID:
- 1 to 12
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Abstract We analyzed 4,754 broadband seismic recordings of the SKS, SKKS, and SPdKS wavefield from 13 high quality events sampling the Samoa ultralow‐velocity zone (ULVZ). We measured differential travel‐times and amplitudes between the SKKS and SKS arrivals, which are highly sensitive to the emergence of the SPdKS seismic phase, which is in turn highly sensitive to lowermost mantle velocity perturbations such as generated by ULVZs. We modeled these data using a 2‐D axi‐symmetric waveform modeling approach and are able to explain these data with a single ULVZ. In order to predict both travel‐time and amplitude perturbations we found that a large ULVZ length in the great circle arc direction on the order of 10° or larger is required. The large ULVZ length limits acceptable ULVZ elastic parameters. Here we find that δVSand δVPreductions from 20% to 22% and 15% to 17% respectively gives us the best fit, with a thickness of 26 km. Initial 3‐D modeling efforts do not recover the extremes in the differential measurements, demonstrating that 3‐D effects are important and must be considered in the future. However, the 3‐D modeling is generally consistent with the velocity reductions recovered with the 2‐D modeling. These velocity reductions are compatible with a compositional component to the ULVZ. Furthermore, geodynamic predictions for a compositional ULVZ that is moving predict a long linear shape similar to the shape of the Samoa ULVZ we confirm in this study.more » « less
-
Abstract Increasing deployment of dense arrays has facilitated detailed structure imaging for tectonic investigation, hazard assessment and resource exploration. Strong velocity heterogeneity and topographic changes have to be considered during passive source imaging. However, it is quite challenging for ray‐based methods, such as Kirchhoff migration or the widely used teleseismic receiver function, to handle these problems. In this study, we propose a 3‐D passive source reverse time migration strategy based on the spectral element method. It is realized by decomposing the time reversal full elastic wavefield into amplitude‐preserved vector P and S wavefields by solving the corresponding weak‐form solutions, followed by a dot‐product imaging condition to get images for the subsurface structures. It enables us to use regional 3‐D migration velocity models and take topographic variations into account, helping us to locate reflectors at more accurate positions than traditional 1‐D model‐based methods, like teleseismic receiver functions. Two synthetic tests are used to demonstrate the advantages of the proposed method to handle topographic variations and complex velocity heterogeneities. Furthermore, applications to the Laramie array data using both teleseismic P and S waves enable us to identify several south‐dipping structures beneath the Laramie basin in southeast Wyoming, which are interpreted as the Cheyenne Belt suture zone and agree with, and improve upon previous geological interpretations.more » « less
-
null (Ed.)Bedload particle hops are defined as successive motions of a particle from start to stop, characterizing one of the most fundamental processes of bedload sediment transport in rivers. Although two transport regimes have been recently identified for short and long hops, respectively, there is still the lack of a theory explaining the mean hop distance–travel time scaling for particles performing short hops, which dominate the transport and may cover over 80 % of the total hop events. In this paper, we propose a velocity-variation-based formulation, the governing equation of which is intrinsically identical to that of Taylor dispersion for solute transport within shear flows. The key parameter, namely the diffusion coefficient, can be determined by hop distances and travel times, which are easier to measure and more accurate than particle accelerations. For the first time, we obtain an analytical solution for the mean hop distance–travel time relation valid for the entire range of travel times, which agrees well with the measured data. Regarding travel times, we identify three distinct regimes in terms of different scaling exponents: respectively, $$\sim$$ 1.5 for the initial regime and $$\sim$$ 5/3 for the transition regime, which define the short hops, and 1 for the Taylor dispersion regime defining long hops. The corresponding distribution of the hop distance is analytically obtained and experimentally verified. We also show that the conventionally used exponential distribution, as proposed by Einstein, is solely for long hops. Further validation of the present formulation is provided by comparing the simulated accelerations with measurements.more » « less
-
The growing number of AI-driven applications in mobile devices has led to solutions that integrate deep learning models with the available edge-cloud resources. Due to multiple benefits such as reduction in on-device energy consumption, improved latency, improved network usage, and certain privacy improvements, split learning, where deep learning models are split away from the mobile device and computed in a distributed manner, has become an extensively explored topic. Incorporating compression-aware methods (where learning adapts to compression level of the communicated data) has made split learning even more advantageous. This method could even offer a viable alternative to traditional methods, such as federated learning techniques. In this work, we develop an adaptive compression-aware split learning method (“deprune”) to improve and train deep learning models so that they are much more network-efficient, which would make them ideal to deploy in weaker devices with the help of edge-cloud resources. This method is also extended (“prune”) to very quickly train deep learning models through a transfer learning approach, which tradesoff little accuracy for much more network-efficient inference abilities. We show that the “deprune” method can reduce network usage by 4× when compared with a split-learning approach (that does not use our method) without loss of accuracy, while also improving accuracy over compression-aware split-learning by up to 4 percent. Lastly, we show that the “prune” method can reduce the training time for certain models by up to 6× without affecting the accuracy when compared against a compression-aware split-learning approach.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.1109/tgrs.2020.3039165
