-
Studying First Passage Problems using Neural Networks: A Case Study in the Slit-Well Microfluidic Device
Authors:
Andrew M. Nagel,
Martin Magill,
Hendrick W. de Haan
Abstract:
This study presents deep neural network solutions to a time-integrated Smoluchowski equation modeling the mean first passage time of nanoparticles traversing the slit-well microfluidic device. This physical scenario is representative of a broader class of parameterized first passage problems in which key output metrics are dictated by a complicated interplay of problem parameters and system geomet…
▽ More
This study presents deep neural network solutions to a time-integrated Smoluchowski equation modeling the mean first passage time of nanoparticles traversing the slit-well microfluidic device. This physical scenario is representative of a broader class of parameterized first passage problems in which key output metrics are dictated by a complicated interplay of problem parameters and system geometry. Specifically, whereas these types of problems are commonly studied using particle simulations of stochastic differential equation models, here the corresponding partial differential equation model is solved using a method based on deep neural networks. The results illustrate that the neural network method is synergistic with the time-integrated Smoluchowski model: together, these are used to construct continuous mappings from key physical inputs (applied voltage and particle diameter) to key output metrics (mean first passage time and effective mobility). In particular, this capability is a unique advantage of the time-integrated Smoluchowski model over the corresponding stochastic differential equation models. Furthermore, the neural network method is demonstrated to easily and reliably handle geometry-modifying parameters, which is generally difficult to accomplish using other methods.
△ Less
Submitted 26 April, 2022;
originally announced April 2022.
-
Neural Network Solutions to Differential Equations in Non-Convex Domains: Solving the Electric Field in the Slit-Well Microfluidic Device
Authors:
Martin Magill,
Andrew M. Nagel,
Hendrick W. de Haan
Abstract:
The neural network method of solving differential equations is used to approximate the electric potential and corresponding electric field in the slit-well microfluidic device. The device's geometry is non-convex, making this a challenging problem to solve using the neural network method. To validate the method, the neural network solutions are compared to a reference solution obtained using the f…
▽ More
The neural network method of solving differential equations is used to approximate the electric potential and corresponding electric field in the slit-well microfluidic device. The device's geometry is non-convex, making this a challenging problem to solve using the neural network method. To validate the method, the neural network solutions are compared to a reference solution obtained using the finite element method. Additional metrics are presented that measure how well the neural networks recover important physical invariants that are not explicitly enforced during training: spatial symmetries and conservation of electric flux. Finally, as an application-specific test of validity, neural network electric fields are incorporated into particle simulations. Conveniently, the same loss functional used to train the neural networks also seems to provide a reliable estimator of the networks' true errors, as measured by any of the metrics considered here. In all metrics, deep neural networks significantly outperform shallow neural networks, even when normalized by computational cost. Altogether, the results suggest that the neural network method can reliably produce solutions of acceptable accuracy for use in subsequent physical computations, such as particle simulations.
△ Less
Submitted 25 April, 2020;
originally announced April 2020.
-
Neural Networks Trained to Solve Differential Equations Learn General Representations
Authors:
Martin Magill,
Faisal Qureshi,
Hendrick W. de Haan
Abstract:
We introduce a technique based on the singular vector canonical correlation analysis (SVCCA) for measuring the generality of neural network layers across a continuously-parametrized set of tasks. We illustrate this method by studying generality in neural networks trained to solve parametrized boundary value problems based on the Poisson partial differential equation. We find that the first hidden…
▽ More
We introduce a technique based on the singular vector canonical correlation analysis (SVCCA) for measuring the generality of neural network layers across a continuously-parametrized set of tasks. We illustrate this method by studying generality in neural networks trained to solve parametrized boundary value problems based on the Poisson partial differential equation. We find that the first hidden layer is general, and that deeper layers are successively more specific. Next, we validate our method against an existing technique that measures layer generality using transfer learning experiments. We find excellent agreement between the two methods, and note that our method is much faster, particularly for continuously-parametrized problems. Finally, we visualize the general representations of the first layers, and interpret them as generalized coordinates over the input domain.
△ Less
Submitted 29 June, 2018;
originally announced July 2018.
-
Waves of DNA: Propagating Excitations in Extended Nanoconfined Polymers
Authors:
Alexander R. Klotz,
Hendrick W. de Haan,
Walter W. Reisner
Abstract:
We use a nanofluidic system to investigate the emergence of thermally driven collective phenomena along a single polymer chain. In our approach, a single DNA molecule is confined in a nanofluidic slit etched with arrays of embedded nanocavities; the cavity lattice is designed so that a single chain occupies multiple cavities. Fluorescent video-microscopy data shows that waves of excess fluorescenc…
▽ More
We use a nanofluidic system to investigate the emergence of thermally driven collective phenomena along a single polymer chain. In our approach, a single DNA molecule is confined in a nanofluidic slit etched with arrays of embedded nanocavities; the cavity lattice is designed so that a single chain occupies multiple cavities. Fluorescent video-microscopy data shows that waves of excess fluorescence propagate across the cavity-straddling molecule, corresponding to propagating fluctuations of contour overdensity in the cavities. The waves are quantified by examining the correlation in intensity fluctuations between neighbouring cavities. Correlations grow from an anti-correlated minimum to a correlated maximum before decaying, corresponding to a transfer of contour between neighbouring cavities at a fixed transfer time-scale. The observed dynamics can be modelled using Langevin dynamics simulations and a minimal lattice model of coupled diffusion. This study shows how confinement-based sculpting of the polymer equilibrium configuration, by renormalizing the physical system into a series of discrete cavity states, can lead to new types of dynamic collective phenomena.
△ Less
Submitted 29 July, 2016; v1 submitted 28 July, 2016;
originally announced July 2016.
-
Simulating the Entropic Collapse of Coarse-Grained Chromosomes
Authors:
Tyler N. Shendruk,
Martin Bertrand,
Hendrick W. de Haan,
James L. Harden,
Gary W. Slater
Abstract:
Depletion forces play a role in the compaction and de-compation of chromosomal material in simple cells but it remains debatable whether they are sufficient to account for chromosomal collapse. We present coarse-grained molecular dynamics simulations, which reveal that depletion-induced attraction is sufficient to cause the collapse of a flexible chain of large structural monomers immersed in a ba…
▽ More
Depletion forces play a role in the compaction and de-compation of chromosomal material in simple cells but it remains debatable whether they are sufficient to account for chromosomal collapse. We present coarse-grained molecular dynamics simulations, which reveal that depletion-induced attraction is sufficient to cause the collapse of a flexible chain of large structural monomers immersed in a bath of smaller depletants. These simulations use an explicit coarse-grained computational model that treats both the supercoiled DNA structural monomers and the smaller protein crowding agents as combinatorial, truncated Lennard-Jones spheres. By presenting a simple theoretical model, we quantitatively cast the action of depletants on supercoiled bacterial DNA as an effective solvent quality. The rapid collapse of the simulated flexible chromosome at the predicted volume fraction of depletants is a continous phase transition. Additional physical effects to such simple chromosome models, such as enthalpic interactions between structural monomers or chain rigidity, are required if the collapse is to be a first-order phase transition.
△ Less
Submitted 16 November, 2014; v1 submitted 10 July, 2014;
originally announced July 2014.
-
Translocation of a Polymer through a Nanopore across a Viscosity Gradient
Authors:
Hendrick W. de Haan,
Gary W. Slater
Abstract:
The translocation of a polymer through a pore in a membrane separating fluids of different viscosities is studied via several computational approaches. Starting with the polymer halfway, we find that as a viscosity difference across the pore is introduced, translocation will predominately occur towards one side of the membrane. These results suggest an intrinsic pumping mechanism for translocation…
▽ More
The translocation of a polymer through a pore in a membrane separating fluids of different viscosities is studied via several computational approaches. Starting with the polymer halfway, we find that as a viscosity difference across the pore is introduced, translocation will predominately occur towards one side of the membrane. These results suggest an intrinsic pumping mechanism for translocation across cell walls which could arise whenever the fluid across the membrane is inhomogeneous. Somewhat surprisingly, the sign of the preferred direction of translocation is found to be strongly dependent on the simulation algorithm: for Langevin Dynamics (LD) simulations, a bias towards the low viscosity side is found while for Brownian Dynamics (BD), a bias towards the high viscosity is found. Examining the translocation dynamics in detail across a wide range of viscosity gradients and developing a simple force model to estimate the magnitude of the bias, the LD results are demonstrated to be more physically realistic. The LD results are also compared to those generated from a simple, one dimensional random walk model of translocation to investigate the role of the internal degrees of freedom of the polymer and the entropic barrier. To conclude, the scaling of the results across different polymer lengths demonstrates the saturation of the preferential direction with polymer length and the non-trivial location of the maximum in the exponent corresponding to the scaling of the translocation time with polymer length.
△ Less
Submitted 14 September, 2012;
originally announced September 2012.
-
Monte Carlo Approaches for Simulating a Particle at a Diffusivity Interface and the "Ito--Stratonovich Dilemma"
Authors:
Hendrick W. de Haan,
Mykyta V. Chubynsky,
Gary W. Slater
Abstract:
The possibility of different interpretations of the stochastic term (or calculi) in the overdamped Langevin equation for the motion of a particle in an inhomogeneous medium is often referred to as the "Ito--Stratonovich dilemma," although there is, in fact, a continuum of choices. We introduce two Monte Carlo (MC) simulation approaches for studying such systems, with both approaches giving the cho…
▽ More
The possibility of different interpretations of the stochastic term (or calculi) in the overdamped Langevin equation for the motion of a particle in an inhomogeneous medium is often referred to as the "Ito--Stratonovich dilemma," although there is, in fact, a continuum of choices. We introduce two Monte Carlo (MC) simulation approaches for studying such systems, with both approaches giving the choice between different calculi (in particular, Ito, Stratonovich, and "isothermal"). To demonstrate these approaches, we study the diffusion on a 1D interval of a particle released at an interface (in the middle of the system) between two media where this particle has different diffusivities (e.g., two fluids with different viscosities). For reflecting boundary conditions at the ends of the interval, a discontinuity at the interface in the stationary-state particle distribution is found, except for the isothermal case, as expected. We also study the first-passage problem using absorbing boundary conditions. Good agreement is found when comparing the MC approaches to theory as well as Brownian and Langevin dynamics simulations. Also, the results themselves turn out to be interesting. For instance: 1) for some calculi, there are more particles on the low-viscosity side (LVS) at earlier times and then fewer particles at later times; 2) there is no preference to end up on a particular wall for the Ito variant, but a bias towards the wall on the LVS in all other cases; 3) the mean first-passage time to the wall on the LVS grows as the viscosity on the other side is increased, except for the isothermal case where it approaches a constant; 4) when the viscosity ratio is high, the first-passage-time distribution for the wall on the LVS is much broader than for the other wall, with a power-law dependence of the former in a certain time interval whose exponent depends on the calculus.
△ Less
Submitted 24 August, 2012;
originally announced August 2012.
