Search for: All records

The semiclassical backreaction equations are solved in closed Robertson-Walker spacetimes containing a positive cosmological constant and a conformally coupled massive scalar field. Renormalization of the stress-energy tensor results in higher derivative terms that can lead to solutions that vary on much shorter time scales than the solutions that would occur if the higher derivative terms were not present. These extra solutions can be eliminated through the use of order reduction. Four different methods of order reduction are investigated. These are first applied to the case when only conformally invariant fields, with and without classical radiation, are present. Then they are applied to the massive conformally coupled scalar field. The effects of different adiabatic vacuum states for the massive field are considered. It is found that if enough particles are produced, then the Universe collapses to a final singularity. Otherwise it undergoes a bounce, but at a smaller value of the scale factor (for the models considered) than occurs for the classical de Sitter solution. The stress-energy tensor incorporates both particle production and vacuum polarization effects. An analysis of the energy density of the massive field is done to determine when the contribution from the particles dominates. 

Abstract. We investigate techniques for using deep neural networks to produce surrogatemodels for short-term climate forecasts. A convolutional neural network istrained on 97 years of monthly precipitation output from the 1pctCO2 run (theCO2 concentration increases by 1 % per year) simulated by the second-generation Canadian Earth System Model (CanESM2). The neural network clearly outperforms a persistence forecast anddoes not show substantially degraded performance even when the forecast lengthis extended to 120 months. The model is prone to underpredicting precipitationin areas characterized by intense precipitation events. Scheduled sampling(forcing the model to gradually use its own past predictions rather than groundtruth) is essential for avoiding amplification of early forecasting errors.However, the use of scheduled sampling also necessitates preforecasting(generating forecasts prior to the first forecast date) to obtain adequateperformance for the first few prediction time steps. We document the trainingprocedures and hyperparameter optimization process for researchers who wish toextend the use of neural networks in developing surrogate models. 

Abstract Simple climate models (SCMs) are computationally efficient and capable of emulating global mean output of more complex Earth system models (ESMs). In doing so, SCMs can play a critical role in climate research as stand‐ins for the computationally more expensive models, especially in studies involving low, spatial, and/or temporal resolution, providing more computationally efficient sources of climate data. Here we use Hector v2.5.0 to emulate the multiforcing historical and RCP scenario output for 31 concentration and seven emission‐driven ESMs. When calibrating Hector, sufficient calibration data must be used to constrain the model; otherwise, climate and/or carbon parameters affecting physical processes may be able to trade off with one another, allowing for solutions to use physically unreasonable fitted parameter values as well as limiting the application of the SCM as an emulator. We also present a novel methodology that uses the ESM range as a calibration data, which can be adopted when faced with missing variable output from a specific model.