arXiv subjects
Publications and source records attributed to A. C. Calder.
We present a look at Ookami, a project providing community access to a testbed supercomputer with the ARM-based A64FX processors developed by a collaboration between RIKEN and Fujitsu and deployed in the Japanese supercomputer Fugaku. We describe the project, provide details about the user base and education/training program, and present highlights from performance studies of two astrophysical simulation codes.
We present examples of validating components of an astrophysical simulation code. Problems of stellar astrophysics are multi-dimensional and involve physics acting on large ranges of length and time scales that are impossible to include in macroscopic models on present computational resources. Simulating these events thus necessitates the development of sub-grid-scale models and the capability to post-process simulations with higher-fidelitymethods.We present an overview of the problem of validating astrophysical models and simulations illustrated with two examples. First, we present a study aimed at validating hydrodynamicswith high energy density laboratory experiments probing shocks and fluid instabilities. Second, we present an effort at validating code modules for use in both macroscopic simulations of astrophysical events and for post processing Lagrangian tracer particles to calculate detailed abundances from thermonuclear reactions occurring during an event.
Our understanding of stars and their fates is based on coupling observations to theoretical models. Unlike laboratory physicists, we cannot perform experiments on stars, but rather must patiently take what nature allows us to observe. Simulation offers a means of virtual experimentation, enabling a detailed understanding of the most violent ongoing explosions in the Universe---the deaths of stars.
We develop an improved method for tracking the nuclear flame during the deflagration phase of a Type Ia supernova, and apply it to study the variation in outcomes expected from the gravitationally confined detonation (GCD) paradigm. A simplified 3-stage burning model and a non-static ash state are integrated with an artificially thickened advection-diffusion-reaction (ADR) flame front in order to provide an accurate but highly efficient representation of the energy release and electron capture in and after the unresolvable flame. We demonstrate that both our ADR and energy release methods do not generate significant acoustic noise, as has been a problem with previous ADR-based schemes. We proceed to model aspects of the deflagration, particularly the role of buoyancy of the hot ash, and find that our methods are reasonably well-behaved with respect to numerical resolution. We show that if a detonation occurs in material swept up by the material ejected by the first rising bubble but gravitationally confined to the white dwarf (WD) surface (the GCD paradigm), the density structure of the WD at detonation is systematically correlated with the distance of the deflagration ignition point from the center of the star. Coupled to a suitably stochastic ignition process, this correlation may provide a plausible explanation for the variety of nickel masses seen in Type Ia Supernovae.
We develop and calibrate a realistic model flame for hydrodynamical simulations of deflagrations in white dwarf (Type Ia) supernovae. Our flame model builds on the advection-diffusion-reaction model of Khokhlov and includes electron screening and Coulomb corrections to the equation of state in a self-consistent way. We calibrate this model flame--its energetics and timescales for energy release and neutronization--with self-heating reaction network calculations that include both these Coulomb effects and up-to-date weak interactions. The burned material evolves post-flame due to both weak interactions and hydrodynamic changes in density and temperature. We develop a scheme to follow the evolution, including neutronization, of the NSE state subsequent to the passage of the flame front. As a result, our model flame is suitable for deflagration simulations over a wide range of initial central densities and can track the temperature and electron fraction of the burned material through the explosion and into the expansion of the ejecta.
We present our first nucleosynthesis results from a numerical simulation of the thermonuclear disruption of a static cold Chandrasekhar-mass C/O white dwarf. The two-dimensional simulation was performed with an adaptive-mesh Eulerian hydrodynamics code, FLASH, that uses as a flame capturing scheme the evolution of a passive scaler. To compute the isotopic yields and their velocity distribution, 10,000 massless tracer particles are embedded in the star. The particles are advected along streamlines and provide a Lagrangian description of the explosion. We briefly describe our verification tests and preliminary results from post-processing the particle trajectories with a modest (214 isotopes) reaction network.
We present the first high-resolution three-dimensional simulations of the deflagration phase of Type Ia supernovae that treat the entire massive white dwarf. We report the results of simulations in which ignition of the nuclear burning occurs slightly off-center. The subsequent evolution of the nuclear burning is surprisingly asymmetric with a growing bubble of hot ash rapidly rising to the stellar surface. Upon reaching the surface, the mass of burned material is $\approx 0.075 M_\sun$ and the kinetic energy is $4.3 \times 10^{49}$ ergs. The velocity of the top of the rising bubble approaches 8000 km s$^{-1}$. The amount of the asymmetry found in the model offers a natural explanation for the observed diversity in Type Ia supernovae. Our study strongly disfavors the classic central-ignition pure deflagration scenario by showing that the result is highly sensitive to details of the initial conditions.
Many classical nova ejecta are enriched in CNO and Ne. Rosner et al. recently suggested that the enrichment might originate in the resonant interaction between large-scale shear flows in the accreted H/He envelope and gravity waves at the interface between the envelope and the underlying C/O white dwarf. The shear flow amplifies the waves, which eventually form cusps and break. This wave breaking injects a spray of C/O into the superincumbent H/He. Using two-dimensional simulations, we formulate a quantitative expression for the amount of C/O per unit area that can be entrained into the H/He at saturation. The fraction of the envelope that is enriched depends on the horizontal distribution of shear velocity and the density contrast between the C/O white dwarf and the H/He layer but is roughly independent of the vertical shape of the shear profile. Using this parameterization for the mixed mass, we then perform several one-dimensional Lagrangian calculations of an accreting white dwarf envelope and consider two scenarios: that the wave breaking and mixing is driven by the convective flows; and that the mixing occurs prior to the onset of convection. In the absence of enrichment prior to ignition, the base of the convective zone, as calculated from mixing-length theory with the Ledoux instability criterion, does not reach the C/O interface. As a result, there is no additional mixing, and the runaway is slow. In contrast, the formation of a mixed layer during the accretion of H/He, prior to ignition, causes a more violent runaway. The envelope can be enriched by <25% of C/O by mass (consistent with that observed in some ejecta) for shear velocities, over the surface, with Mach numbers <0.4.
Recent Chandra and XMM-Newton observations of galaxy cluster cooling flows have revealed X-ray emission voids of up to 30 kpc in size that have been identified with buoyant, magnetized bubbles. Motivated by these observations, we have investigated the behavior of rising bubbles in stratified atmospheres using the FLASH adaptive-mesh simulation code. We present results from 2-D simulations with and without the effects of magnetic fields, and with varying bubble sizes and background stratifications. We find purely hydrodynamic bubbles to be unstable; a dynamically important magnetic field is required to maintain a bubble's integrity. This suggests that, even absent thermal conduction, for bubbles to be persistent enough to be regularly observed, they must be supported in large part by magnetic fields. Thermal conduction unmitigated by magnetic fields can dissipate the bubbles even faster. We also observe that the bubbles leave a tail as they rise; the structure of these tails can indicate the history of the dynamics of the rising bubble.
Critically understanding the `standard candle'-like behavior of Type Ia supernovae requires understanding their explosion mechanism. One family of models for Type Ia Supernovae begins with a deflagration in a Carbon-Oxygen white dwarf which greatly accelerates through wrinkling and flame instabilities. While the planar speed and behavior of astrophysically-relevant flames is increasingly well understood, more complex behavior, such as the flame's response to stretch and curvature, has not been extensively explored in the astrophysical literature; this behavior can greatly enhance or suppress instabilities and local flame-wrinkling, which in turn can increase or decrease the bulk burning rate. In this paper, we explore the effects of curvature on both nuclear flames and simpler model flames to understand the effect of curvature on the flame structure and speed.
We look in detail at the process of mapping an astrophysical initial model from a stellar evolution code onto the computational grid of an explicit, Godunov type code while maintaining hydrostatic equilibrium. This mapping process is common in astrophysical simulations, when it is necessary to follow short-timescale dynamics after a period of long timescale buildup. We look at the effects of spatial resolution, boundary conditions, the treatment of the gravitational source terms in the hydrodynamics solver, and the initialization process itself. We conclude with a summary detailing the mapping process that yields the lowest ambient velocities in the mapped model.
Observed novae abundances and explosion energies estimated from observations indicate that there must be significant mixing of the heavier material of the white dwarf (C+O) into the lighter accreted material (H+He). Accordingly, nova models must incorporate a mechanism that will dredge up the heavier white dwarf material, and fluid motions from an early convection phase is one proposed mechanism. We present results from two-dimensional simulations of classical nova precursor models that demonstrate the beginning of a convective phase during the `simmering' of a Nova precursor. We use a new hydrostatic equilibrium hydrodynamics module recently developed for the adaptive-mesh code FLASH. The two-dimensional models are based on the one-dimensional models of Ami Glasner, and were evolved with FLASH from a pre-convective state to the onset of convection.
We present the results of a simulation of a wind-driven non-linear gravity wave breaking on the surface of a white dwarf. The ``wind'' consists of H/He from an accreted envelope, and the simulation demonstrates that this breaking wave mechanism can produce a well-mixed layer of H/He with C/O from the white dwarf above the surface. Material from this mixed layer may then be transported throughout the accreted envelope by convection, which would enrich the C/O abundance of the envelope as is expected from observations of novae.
We present a case study of validating an astrophysical simulation code. Our study focuses on validating FLASH, a parallel, adaptive-mesh hydrodynamics code for studying the compressible, reactive flows found in many astrophysical environments. We describe the astrophysics problems of interest and the challenges associated with simulating these problems. We describe methodology and discuss solutions to difficulties encountered in verification and validation. We describe verification tests regularly administered to the code, present the results of new verification tests, and outline a method for testing general equations of state. We present the results of two validation tests in which we compared simulations to experimental data. The first is of a laser-driven shock propagating through a multi-layer target, a configuration subject to both Rayleigh-Taylor and Richtmyer-Meshkov instabilities. The second test is a classic Rayleigh-Taylor instability, where a heavy fluid is supported against the force of gravity by a light fluid. Our simulations of the multi-layer target experiments showed good agreement with the experimental results, but our simulations of the Rayleigh-Taylor instability did not agree well with the experimental results. We discuss our findings and present results of additional simulations undertaken to further investigate the Rayleigh-Taylor instability.
We present direct numerical simulations of mixing at Rayleigh-Taylor unstable interfaces performed with the FLASH code, developed at the ASCI/Alliances Center for Astrophysical Thermonuclear Flashes at the University of Chicago. We present initial results of single-mode studies in two and three dimensions. Our results indicate that three-dimensional instabilities grow significantly faster than two-dimensional instabilities and that grid resolution can have a significant effect on instability growth rates. We also find that unphysical diffusive mixing occurs at the fluid interface, particularly in poorly resolved simulations.
We discuss some of the computational challenges encountered in simulating the evolution of clusters of galaxies. Eulerian adaptive mesh refinement (AMR) techniques can successfully address these challenges but are currently being used by only a few groups. We describe our publicly available AMR code, FLASH, which uses an object-oriented framework to manage its AMR library, physics modules, and automated verification. We outline the development of the FLASH framework to include collisionless particles, permitting it to be used for cluster simulation.
We couple two-dimensional hydrodynamics to realistic one-dimensional multigroup flux-limited diffusion neutrino transport to investigate protoneutron star convection in core collapse supernovae, and more specifically, the interplay between its development and neutrino transport, for both 15 and 25 solar mass models. In the presence of neutrino transport, protoneutron star convection velocities are too small relative to bulk inflow velocities to result in any significant convective transport of entropy and leptons. A simple analytical model supports our numerical results, indicating that the inclusion of neutrino transport reduces the entropy-driven (lepton-driven) convection growth rates and asymptotic velocities by a factor of 3 (50) at the neutrinosphere and a factor 250 (1000) at a density of 10^{12} g/cm^{3}, for both our 15 and 25 solar mass models. Moreover, when transport is included, the initial postbounce entropy gradient is smoothed out by neutrino diffusion, whereas the initial lepton gradient is maintained by electron capture and neutrino escape near the neutrinosphere. Despite the maintenance of the lepton gradient, protoneutron star convection does not develop over the 100 ms duration typical of all our simulations, except in the instance where ``low-test'' initial conditions are used, which are generated by core collapse and bounce simulations that neglect neutrino-electron scattering and ion-ion screening corrections to neutrino- nucleus elastic scattering.
We investigate neutrino-driven convection in core collapse supernovae and its ramifications for the explosion mechanism, for a 15 solar mass model. Our two-dimensional simulation begins at 12 ms after bounce and proceeds for 500 ms. We couple two-dimensional PPM hydrodynamics to precalculated one-dimensional MGFLD neutrino transport. (The accuracy of this approximation is assessed.) For the moment we sacrifice dimensionality for realism in other aspects of our neutrino transport. MGFLD is an implementation of neutrino transport that simultaneously (a) is multigroup and (b) simulates with greater realism the transport of neutrinos in opaque, semitransparent, and transparent regions. Both are crucial to the accurate determination of postshock neutrino heating, which sensitively depends on the luminosities, spectra, and flux factors of the electron neutrinos and antineutrinos emerging from their respective neutrinospheres. By 212 ms after bounce, we see large-scale, neutrino-driven convection beneath the shock, characterized by higher-entropy, expanding upflows and denser, lower-entropy, finger-like downflows. The radial convection velocities at this time become supersonic just below the shock, reaching magnitudes in excess of 10^{9} cm/sec. Eventually, however, the shock recedes to smaller radii, and at 500 ms after bounce there is no evidence in our simulation of an explosion. While vigorous, neutrino-driven convection in our model does not have a significant impact on the overall shock dynamics. The differences between our results and those of other groups are considered. These most likely result from differences in (1) numerical hydrodynamics methods, (2) initial postbounce models, and most important, (3) neutrino transport approximations.