-
Reflection on modern methods: Good practices for applied statistical learning in epidemiology
Authors:
Yanelli Nunez,
Elizabeth A. Gibson,
Eva M. Tanner,
Chris Gennings,
Brent A. Coull,
Jeff A. Goldsmith,
Marianthi-Anna Kioumourtzoglou
Abstract:
Statistical learning (SL) includes methods that extract knowledge from complex data. SL methods beyond generalized linear models are being increasingly implemented in public health research and epidemiology because they can perform better in instances with complex or high-dimensional data---settings when traditional statistical methods fail. These novel methods, however, often include random sampl…
▽ More
Statistical learning (SL) includes methods that extract knowledge from complex data. SL methods beyond generalized linear models are being increasingly implemented in public health research and epidemiology because they can perform better in instances with complex or high-dimensional data---settings when traditional statistical methods fail. These novel methods, however, often include random sampling which may induce variability in results. Best practices in data science can help to ensure robustness. As a case study, we included four SL models that have been applied previously to analyze the relationship between environmental mixtures and health outcomes. We ran each model across 100 initializing values for random number generation, or "seeds," and assessed variability in resulting estimation and inference. All methods exhibited some seed-dependent variability in results. The degree of variability differed across methods and exposure of interest. Any SL method reliant on a random seed will exhibit some degree of seed sensitivity. We recommend that researchers repeat their analysis with various seeds as a sensitivity analysis when implementing these methods to enhance interpretability and robustness of results.
△ Less
Submitted 2 October, 2020; v1 submitted 12 June, 2020;
originally announced June 2020.
-
The isothermal evolution of a shock-filament interaction
Authors:
K. J. A. Goldsmith,
J. M. Pittard
Abstract:
Studies of filamentary structures that are prevalent throughout the interstellar medium are of great significance to a number of astrophysical fields. Here, we present 3D hydrodynamic simulations of shock-filament interactions where the equation of state has been softened to become almost isothermal. We investigate the effect of such an isothermal regime on the interaction (where both the shock an…
▽ More
Studies of filamentary structures that are prevalent throughout the interstellar medium are of great significance to a number of astrophysical fields. Here, we present 3D hydrodynamic simulations of shock-filament interactions where the equation of state has been softened to become almost isothermal. We investigate the effect of such an isothermal regime on the interaction (where both the shock and filament are isothermal), and we examine how the nature of the interaction changes when the orientation of the filament, the shock Mach number, and the filament density contrast are varied. We find that only sideways-oriented filaments with a density contrast of $10^2$ form a three-rolled structure, dissimilar to the results of a previous study. Moreover, the angle of orientation of the filament plays a large role in the evolution of the filament morphology: the greater the angle of orientation, the longer and less turbulent the wake. Turbulent stripping of filament material leading to fragmentation of the core occurs in most filaments; however, filaments orientated at an angle of $85^{\circ}$ to the shock front do not fragment and are longer-lived. In addition, values of the drag time are influenced by the filament length, with longer filaments being accelerated faster than shorter ones. Furthermore, filaments in an isothermal regime exhibit faster acceleration than those struck by an adiabatic shock. Finally, we find that the drag and mixing times of the filament increase as the angle of orientation of the filament is increased.
△ Less
Submitted 11 December, 2019;
originally announced December 2019.
-
A comparison of shock-cloud and wind-cloud interactions: effect of increased cloud density contrast on cloud evolution
Authors:
K. J. A. Goldsmith,
J. M. Pittard
Abstract:
The similarities, or otherwise, of a shock or wind interacting with a cloud of density contras t$ χ= 10$ were explored in a previous paper. Here, we investigate such interactions with clouds of higher density contrast. We compare the adiabatic hydrodynamic interaction of a Mach 10 shock with a spherical cloud of $χ= 10^{3}$ with that of a cloud embedded in a wind with identical parameters to the p…
▽ More
The similarities, or otherwise, of a shock or wind interacting with a cloud of density contras t$ χ= 10$ were explored in a previous paper. Here, we investigate such interactions with clouds of higher density contrast. We compare the adiabatic hydrodynamic interaction of a Mach 10 shock with a spherical cloud of $χ= 10^{3}$ with that of a cloud embedded in a wind with identical parameters to the post-shock flow. We find that initially there are only minor morphological differences between the shock-cloud and wind-cloud interactions, compared to when $χ= 10$. However, once the transmitted shock exits the cloud, the development of a turbulent wake and fragmentation of the cloud differs between the two simulations. On increasing the wind Mach number, we note the development of a thin, smooth tail of cloud material, which is then disrupted by the fragmentation of the cloud core and subsequent `mass-loading' of the flow. We find that the normalized cloud mixing time ($t_{mix}$) is shorter at higher $χ$. However, a strong Mach number dependence on tmix and the normalized cloud drag time, $t'_{drag}$, is not observed. Mach-number-dependent values of $t_{mix}$ and $t'_{drag}$ from comparable shock-cloud interactions converge towards the Mach-number-independent time-scales of the wind-cloud simulations. We find that high $χ$ clouds can be accelerated up to 80-90 per cent of the wind velocity and travel large distances before being significantly mixed. However, complete mixing is not achieved in our simulations and at late times the flow remains perturbed.
△ Less
Submitted 12 May, 2018;
originally announced May 2018.
-
A comparison of shock-cloud and wind-cloud interactions: The longer survival of clouds in winds
Authors:
K. J. A. Goldsmith,
J. M. Pittard
Abstract:
The interaction of a hot, high-velocity wind with a cold, dense molecular cloud has often been assumed to resemble the evolution of a cloud embedded in a post-shock flow. However, no direct comparative study of these two processes currently exists in the literature. We present 2D adiabatic hydrodynamical simulations of the interaction of a Mach 10 shock with a cloud of density contrast $χ= 10$ and…
▽ More
The interaction of a hot, high-velocity wind with a cold, dense molecular cloud has often been assumed to resemble the evolution of a cloud embedded in a post-shock flow. However, no direct comparative study of these two processes currently exists in the literature. We present 2D adiabatic hydrodynamical simulations of the interaction of a Mach 10 shock with a cloud of density contrast $χ= 10$ and compare our results with those of a commensurate wind-cloud simulation. We then investigate the effect of varying the wind velocity, effectively altering the wind Mach number $M_{wind}$, on the cloud's evolution. We find that there are significant differences between the two processes: 1) the transmitted shock is much flatter in the shock-cloud interaction; 2) a low-pressure region in the wind-cloud case deflects the flow around the edge of the cloud in a different manner to the shock-cloud case; 3) there is far more axial compression of the cloud in the case of the shock. As $M_{wind}$ increases, the normalised rate of mixing is reduced. Clouds in winds with higher $M_{wind}$ also do not experience a transmitted shock through the cloud's rear and are more compressed axially. In contrast with shock-cloud simulations, the cloud mixing time normalised by the cloud-crushing time-scale $t_{cc}$ increases for increasing $M_{wind}$ until it plateaus (at $t_{mix} \simeq 25 \, t_{cc}$) at high $M_{wind}$, thus demonstrating the expected Mach scaling. In addition, clouds in high Mach number winds are able to survive for long durations and are capable of being moved considerable distances.
△ Less
Submitted 12 June, 2017;
originally announced June 2017.
-
The interaction of a magnetohydrodynamical shock with a filament
Authors:
K. J. A. Goldsmith,
J. M. Pittard
Abstract:
We present 3D magnetohydrodynamic numerical simulations of the adiabatic interaction of a shock with a dense, filamentary cloud. We investigate the effects of various filament lengths and orientations on the interaction using different orientations of the magnetic field, and vary the Mach number of the shock, the density contrast of the filament, and the plasma beta, in order to determine their ef…
▽ More
We present 3D magnetohydrodynamic numerical simulations of the adiabatic interaction of a shock with a dense, filamentary cloud. We investigate the effects of various filament lengths and orientations on the interaction using different orientations of the magnetic field, and vary the Mach number of the shock, the density contrast of the filament, and the plasma beta, in order to determine their effect on the evolution and lifetime of the filament. We find that in a parallel magnetic field filaments have longer lifetimes if they are orientated more 'broadside' to the shock front, and that an increase in the density contrast hastens the destruction of the cloud, in terms of the modified cloud-crushing time-scale, tcs. The combination of a mild shock and a perpendicular or oblique field provides the best condition for extending the life of the filament, with some filaments able to survive almost indefinitely since they are cocooned by the magnetic field. A high value for the density contrast does not initiate large turbulent instabilities in either the perpendicular or oblique field cases but rather draws the filament out into long tendrils which may eventually fragment. In addition, flux ropes are only formed in parallel magnetic fields. The length of the filament is, however, not as important for the evolution and destruction of a filament.
△ Less
Submitted 13 July, 2016;
originally announced July 2016.
-
Numerical Simulations of a Shock-Filament Interaction
Authors:
J. M. Pittard,
K. J. A. Goldsmith
Abstract:
We present 3D hydrodynamic adiabatic simulations of a shock interacting with a dense, elongated cloud. We compare how the nature of the interaction changes with the filament's length and its orientation to the shock, and with the shock Mach number and the density contrast of the filament. We then examine the differences with respect to 3D spherical-cloud calculations. We find significant differenc…
▽ More
We present 3D hydrodynamic adiabatic simulations of a shock interacting with a dense, elongated cloud. We compare how the nature of the interaction changes with the filament's length and its orientation to the shock, and with the shock Mach number and the density contrast of the filament. We then examine the differences with respect to 3D spherical-cloud calculations. We find significant differences in the morphology of the interaction when M=10 and chi=100: in many cases 3 parallel rolls are formed, and spread further apart with time, and periodic vortex shedding can occur off the ends of oblique filaments. Sideways-on filaments are accelerated more quickly, and initially lose mass more quickly than spherical clouds due to their greater surface area to volume ratio. However, at late stages they lose mass more slowly, due to the reduced relative speed between the filament and the postshock flow. The acceleration and mixing timescales can vary by a factor of 2 as the filament orientation changes. Oblique filaments can achieve transverse velocities up to 10% of the shock speed. Some aspects of our simulations are compared against experimental and numerical work on rigid cylinders.
△ Less
Submitted 11 February, 2016; v1 submitted 18 December, 2015;
originally announced December 2015.
