Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Linearization error”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,729 records · Page 96

Essay: Bob Siemann-SLC Days at SLAC

Bob Siemann was a great experimentalist and an excellent teacher.We will greatly miss him. Bob came to SLAC in early 1991 to work on the Stanford Linear Collider (SLC). The SLC was a challenging accelerator which began operating in the late 1980's but still had numerous obstacles to be overcome years into operation. One of the compounding difficulties was making reproducible measurements, since the stability of the collider was poor and the diagnostics were insufficient. Bob dove into this challenge and helped design experiments and diagnostics that provided further clarity. I first got to know Bob while I was still a graduate student, trying to finish my thesis and performing some experimental studies on the SLC, which, at the time, was proving to be very difficult. Most of my expertise had been in beam theory and simulation. Dealing with the real issues of the accelerator was challenging. Bob helped me understand the difference between systematic and statistical errors, and separate operational issues from the fundamental physics. His way of teaching was not to provide an explanation but to ask enough questions so that I could find the answer on my own - this was the best way to learn. I later asked Bob to be a reader on my thesis. As in all things, he took this role extremely seriously. He read through the draft and marked every page to the point where I was regretting my decision. However, his questions again helped me understand my own work better and greatly improved my thesis. Bob was also the de facto leader of an effort focused on the damping rings and the bunch compressors. He was great to work with. He made people think for themselves and refused to simply provide answers. He also worked hard himself, expressing real interest and curiosity. After the studies of the SLC damping rings identified a sawtooth instability due to the vacuum chamber impedance as a source of many downstream fluctuations, Bob took charge of upgrading the rings. As part of this program, I suggested an extensive upgrade that also replaced the dipoles with combined function magnets which might have reduced the horizontal emittance another factor of 3. Although he was extremely busy, Bob helped me develop the proposal and understand the magnetic limitations as well as the potential impacts on the beam dynamics. He helped me consider issues well beyond my initial scope. While the proposal never went anywhere and I think Bob had been aware that there was no funding to pursue the option, he saw that it would be a great learning experience for me and it was. In the early 1990's I had simulated a new regime for the beam-ion instability and, with Frank Zimmermann, I developed a model for the effect which was predicted to occur within the high current, low emittance bunch trains in future storage rings or linear colliders. I thought this was pretty good work but Bob convinced me that the next step had to be confirming the theory with measurements. Because the growth rate was inversely dependent on beam sizes and proportional to the vacuum pressure, measurements required significantly increasing the vacuum pressure in existing facilities. Most people discounted trying such an experiment, but with Bob's urging and suggestions and John Byrd's excitement, we managed to make the measurements at the Advanced Light Source (ALS) at Berkeley. By the mid-1990's Bob was completely focused on advanced acceleration concepts and I was not interacting with him as often. At the time, SLAC was putting together a large effort in designing and documenting a design for the Next Linear Collider (NLC) while constructing the NLC Test Accelerator. Bob was worried that a straightforward extrapolation of the microwave technology would be difficult to bring to fruition because of the cost. He wanted to focus on more cost-effective approaches that could enable future accelerators for high energy physics. As usual, he was correct. The experimental programs that he started in direct laser acceleration and plasma-wakefield acceleration have made great progress. He accomplished this with lots of hard work and by engaging the people around him, especially students and postdocs. In the process, he created a group of extremely talented people which has enabled these technologies to be developed to the point where it seems likely that they are viable and will offer two cost-effective approaches to high-gradient acceleration.

43 PARTICLE ACCELERATORS↗

A Scalable Reduced‐Order Model for the Steady Navier–Stokes Equations

Scaling up new scientific technologies from laboratory to industry often involves demonstrating performance on a larger scale. Computer simulations can accelerate design and predictions in the deployment process, though traditional numerical methods are computationally intractable even for intermediate pilot plant scales. Recently, the component reduced order modeling method has been developed to tackle this challenge by combining projection reduced order modeling and discontinuous Galerkin domain decomposition. However, while many scientific or engineering applications involve nonlinear physics, this method has only been demonstrated for various linear systems. In this work, the component reduced order modeling method is extended to steady Navier–Stokes flow, with application to general nonlinear physics in view. The large‐scale, global domain is decomposed into a combination of small‐scale unit component. Linear subspaces for flow velocity and pressure are identified via proper orthogonal decomposition over sample snapshots collected from each small‐scale unit component. Velocity bases are augmented with a pressure supremizer to satisfy the inf–sup condition for stable pressure prediction. Two different nonlinear reduced order modeling methods are employed and compared for efficient evaluation of nonlinear advection: A third‐order tensor projection operator and the empirical quadrature procedure. The proposed method is demonstrated on the flow over arrays of five different unit objects, achieving a 23‐fold speedup with less than 4% relative error in domains up to 256 times larger than the unit components. Furthermore, a numerical experiment with the pressure supremizer strongly indicates the need for a supremizer for stable pressure prediction. A comparison between the tensorial approach and the empirical quadrature procedure revealed a slight advantage of the empirical quadrature procedure. The framework is compared with an alternating Schwarz‐based reduced‐order approach, demonstrating improved efficiency and robustness for the DG‐based global solver while retaining flexibility for sub‐scale iterative solvers. The method is further extended to a coupled advection–diffusion and Navier–Stokes system, illustrating its applicability to multi‐physics problems and its potential for more general, inter‐coupled nonlinear systems.

42 ENGINEERING↗

Multiresolution quantum chemistry in multiwavelet bases: excited states from time-dependent Hartree–Fock and density functional theory via linear response

Using the fully numerical method for time-dependent Hartree–Fock and density functional theory (TD-HF/DFT) with the Tamm–Dancoff (TD) approximation we use a multiresolution analysis (MRA) approach to present our findings. From a reformulation with effective use of the density matrix operator, we obtain a general form of the HF/DFT linear response equation in the first quantization formalism. It can be readily rewritten as an integral equation with the bound-state Helmholtz (BSH) kernel for the Green's function. The MRA implementation of the resultant equation permits excited state calculations without virtual orbitals. Moreover, the integral equation is efficiently and adaptively solved using a numerical multiresolution solver with multiwavelet bases. Our implementation of the TD-HF/DFT methods is applied for calculating the excitation energies of H 2 , Be, N 2 , H 2 O, and C 2 H 4 molecules. The numerical errors of the calculated excitation energies converge in proportion to the residuals of the equation in the molecular orbitals and response functions. The energies of the excited states at a variety of length scales ranging from short-range valence excitations to long-range Rydberg-type ones are consistently accurate. It is shown that the multiresolution calculations yield the correct exponential asymptotic tails for the response functions, whereas those computed with Gaussian basis functions are too diffuse or decay too rapidly. Finally, we introduce a simple asymptotic correction to the local spin-density approximation (LSDA) so that in the TDDFT calculations, the excited states are correctly bound.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Wave-by-Wave Forecasting via Assimilation of Marine Radar Data

Abstract This work describes a phase-resolving wave-forecasting algorithm that is based on the assimilation of marine radar image time series. The algorithm is tested against synthetic data and field observations. The algorithm couples X-band marine radar observations with a phase-resolving wave model that uses the linear mild slope equations for reconstruction of water surface elevations over a large domain of O (km) and a prescribed time window of O (min). The reconstruction also enables wave-by-wave forecasting through forward propagation in space and time. Marine radar image time series provide the input wave observations through a previously given relationship between backscatter intensity and the radial component of the sea surface slope. The algorithm assimilates the wave slope information into the model via a best-fit wave source function at the boundary that minimizes the slope reconstruction error over an annular region at the outer ranges of the radar images. The wave model is then able to propagate the waves across a polar domain to a location of interest at nearer ranges. The constraints on the method for achieving real-time forecasting are identified, and the algorithm is verified against synthetic data and tested using field observations.

Engineering↗

Replication Concepts for Bioenergy Research Experiments

While there are some large and fundamental differences among disciplines related to the conversion of biomass to bioenergy, all scientific endeavors involve the use of bio- logical feedstocks. As such, nearly every scientific experiment conducted in this area, regardless of the specific discipline, is subject to random variation, some of which is unpredictable and unidentifiable (i.e., pure random variation such as variation among plots in an experiment, individuals within a plot, or laboratory samples within an experimental unit) while some is predictable and identifiable (repeatable variation, such as spatial or temporal patterns within an experimental field, a glasshouse or growth chamber, or among laboratory containers). Identifying the scale and sources of this variation relative to the specific hypotheses of interest is a critical component of designing good experiments that generate meaningful and believable hypothesis tests and inference statements. Many bioenergy feedstock experiments are replicated at an incorrect scale, typically by sampling feedstocks to estimate laboratory error or by completely ignoring the errors associated with growing feedstocks in an agricultural area at a field or farmland (micro- or macro-region) scale. As such, actual random errors inherent in experimental materials are frequently underestimated, with unrealistically low standard errors of statistical parameters (e.g., means), leading to improper inferences. The examples and guidelines set forth in this paper and many of the references cited are intended to form the general policy and guidelines for replication of bioenergy feedstock experiments to be published in BioEnergy Research .

59 BASIC BIOLOGICAL SCIENCES↗

Earth's Magnetic Field Measurements for the LCLS Undulators

Measurements of the earth's magnetic field at several locations at SLAC were conducted to determine the possible field error contribution from tuning the undulators in a location with a different magnetic field than that which will be found in the undulator hall. An average difference of 0.08 {+-} 0.04 Gauss has been measured between the downward earth's field components in the test facility and SLAC tunnel locations.

43 PARTICLE ACCELERATORS↗

One-point remapping of Lagrangian perturbation theory in the mildly non-linear regime of cosmic structure formation

On the smallest scales, three-dimensional large-scale structure surveys contain a wealth of cosmological information which cannot be trivially extracted due to the non-linear dynamical evolution of the density field. Lagrangian perturbation theory (LPT) is widely applied to the generation of mock halo catalogs and data analysis. In this work, we compare topological features of the cosmic web such as voids, sheets, filaments and clusters, in the density fields predicted by LPT and full numerical simulation of gravitational large-scale structure formation. We propose a method designed to improve the correspondence between these density fields, in the mildly non-linear regime. We develop a computationally fast and flexible tool for a variety of cosmological applications. Our method is based on a remapping of the approximately-evolved density field, using information extracted from N-body simulations. The remapping procedure consists of replacing the one-point distribution of the density contrast by one which better accounts for the full gravitational dynamics. As a result, we obtain a physically more pertinent density field on a point-by-point basis, while also improving higher-order statistics predicted by LPT. We quantify the approximation error in the power spectrum and in the bispectrum as a function of scale and redshift. Our remapping procedure improves one-, two- and three-point statistics at scales down to 8 Mpc/h.

79 ASTRONOMY AND ASTROPHYSICS↗

Continuous-variable quantum computation of the O(3) model in 1+1 dimensions

We formulate the $O(3)$ non-linear sigma model in $1+1$ dimensions as a limit of a three-component scalar field theory restricted to the unit sphere in the large squeezing limit. This allows us to describe the model in terms of the continuous variable (CV) approach to quantum computing. Here we construct the ground state and excited states using the coupled cluster ansatz and find excellent agreement with the exact diagonalization results for a small number of lattice sites. We then present the simulation protocol for the time evolution of the model using CV gates, estimate the discretization error, and present numerical results obtained from a photonic quantum simulator. We expect that the methods developed in this work will be useful for exploring interesting dynamics for a wide class of sigma models and gauge theories, as well as for simulating scattering events on quantum hardware in the coming decade.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

SU-D-18C-05: Variable Bolus Arterial Spin Labeling MRI for Accurate Cerebral Blood Flow and Arterial Transit Time Mapping

Purpose: Arterial spin labeling (ASL) is an MRI perfusion imaging method from which quantitative cerebral blood flow (CBF) maps can be calculated. Acquisition with variable post-labeling delays (PLD) and variable TRs allows for arterial transit time (ATT) mapping and leads to more accurate CBF quantification with a scan time saving of 48%. In addition, T1 and M0 maps can be obtained without a separate scan. In order to accurately estimate ATT and T1 of brain tissue from the ASL data, variable labeling durations were invented, entitled variable-bolus ASL. Methods: All images were collected on a healthy subject with a 3T Siemens Skyra scanner. Variable-bolus Psuedo-continuous ASL (PCASL) images were collected with 7 TI times ranging 100-4300ms in increments of 700ms with TR ranging 1000-5200ms. All boluses were 1600ms when the TI allowed, otherwise the bolus duration was 100ms shorter than the TI. All TI times were interleaved to reduce sensitivity to motion. Voxel-wise T1 and M0 maps were estimated using a linear least squares fitting routine from the average singal from each TI time. Then pairwise subtraction of each label/control pair and averaging for each TI time was performed. CBF and ATT maps were created using the standard model by Buxton et al. with a nonlinear fitting routine using the T1 tissue map. Results: CBF maps insensitive to ATT were produced along with ATT maps. Both maps show patterns and averages consistent with literature. The T1 map also shows typical T1 contrast. Conclusion: It has been demonstrated that variablebolus ASL produces CBF maps free from the errors due to ATT and tissue T1 variations and provides M0, T1, and ATT maps which have potential utility. This is accomplished with a single scan in a feasible scan time (under 6 minutes) with low sensivity to motion.

60 APPLIED LIFE SCIENCES↗

SU-E-CAMPUS-T-04: Measurement of Proton Pencil Beam Spot Profile Using Cherenkov Radiation in Two Dimensional Optical Fiber Arrays

Purpose: Proton therapy aims to deliver a high dose in a well-defined target volume while sparing the healthy surrounding tissues thanks to their inherent depth dose characteristic (Bragg peak). In proton therapy, several techniques can be used to deliver the dose into the target volume. The one that allows the best conformity with the tumor, is called PBS (Pencil Beam Scanning). The measurement of the proton pencil beam spot profile (spot size) and position is very important for the accurate delivery of dose to the target volume with a good conformity. Methods: We have developed a fine segmented detector array to monitor the PBS. A prototype beam monitor using Cherenkov radiation in clear plastic optical fibers (cPOF) has been developed for continuous display of the pencil beam status during the therapeutic proton Pencil Beam Scanning mode operation. The benefit of using Cherenkov radiation is that the optical output is linear to the dose. Pedestal substraction and the gain adjustment between channels are performed. Spot profiles of various pencil beam energies(100 MeV to 226 MeV) are measured. Two dimensional gaussian fit is used to analyze the beam width and the spot center. The results are compared with that of Lynx(Scintillator-based sensor with CCD camera) and EBT3 Film. Results: The measured gaussian widths using fiber array system changes from 13 to 5 mm for the beam energies from 100 to 226 MeV. The results agree well with Lynx and Film within the systematic error. Conclusion: The results demonstrate good monitoring capability of the system. Not only measuing the spot profile but also monitoring dose map by accumulating each spot measurement is available. The x-y monitoing system with 128 channel readout will be mounted to the snout for the in-situ real time monitoring.

07 ISOTOPE AND RADIATION SOURCES↗

MIONet: Learning Multiple-Input Operators via Tensor Product

As an emerging paradigm in scientific machine learning, neural operators aim to learn operators, via neural networks, that map between infinite-dimensional function spaces. Several neural operators have been recently developed. However, all the existing neural operators are only designed to learn operators defined on a single Banach space; i.e., the input of the operator is a single function. Here, for the first time, we study the operator regression via neural networks for multiple-input operators defined on the product of Banach spaces. We first prove a universal approximation theorem of continuous multiple-input operators. We also provide a detailed theoretical analysis including the approximation error, which provides guidance for the design of the network architecture. Based on our theory and a low-rank approximation, we propose a novel neural operator, MIONet, to learn multiple-input operators. MIONet consists of several branch nets for encoding the input functions and a trunk net for encoding the domain of the output function. Here, we demonstrate that MIONet can learn solution operators involving systems governed by ordinary and partial differential equations. In our computational examples, we also show that we can endow MIONet with prior knowledge of the underlying system, such as linearity and periodicity, to further improve accuracy.

97 MATHEMATICS AND COMPUTING↗

Comparison of Spatial Discretization Methods for Solving the S N Equations Using a Three-Dimensional Method of Manufactured Solutions Benchmark Suite with Escalating Order of Nonsmoothness

For this study, a comparison of the accuracy and computational efficiency of spatial discretization methods of the three-dimensional S N equations is conducted, including discontinuous Galerkin finite element methods, the arbitrarily high-order transport method of nodal type (AHOTN), the linear-linear method, the linear-nodal (LN) method, and the higher-order diamond difference method. For this purpose, we have developed a suite of method of manufactured solutions benchmarks that provides an exact solution of the S N equations even in the presence of scattering. Most importantly, our benchmark suite permits the user to set an arbitrary level of smoothness of the exact solution across the singular characteristics. Our study focuses on the computational efficiency of the considered spatial discretization methods. Numerical results indicate that the best-performing method depends on the norm used to measure the discretization error. We employ discrete Lp norms and integral error norms in this work. For configurations with continuous exact angular flux, high-order AHOTNs perform best under L p error norms, while the LN method performs best when measured by integral error norms. Lastly, when the angular flux is discontinuous, a new singular-characteristic tracking method for three-dimensional geometries performs best among the considered methods.

97 MATHEMATICS AND COMPUTING↗

Prospects for the study of the {tau}-tilde system in SPS1a' at the ILC

The prospects for the analysis of {tau} channels at the supersymmetry (SUSY) benchmark point SPS1a', especially from e{sup +}e{sup -{yields}{tau}}-tilde{sub 1}{sup +{tau}}-tilde{sub 1}{sup -} and e{sup +}e{sup -{yields}{tau}}-tilde{sub 2}{sup +{tau}}-tilde{sub 2}{sup -}, have been studied in full simulation of the ILD detector concept foreseen for the International Linear Collider. All accessible SUSY channels as well as all standard model backgrounds were generated at a center-of-mass energy E{sub CMS}=500 GeV, including the beam energy spectrum and beam backgrounds expected for nominal ILC beam parameters. With an integrated luminosity of 500 fb{sup -1} delivered to the experiment, the masses of the {tau}-tilde{sub 1} and {tau}-tilde{sub 2} can be measured to 107.73{sub -0.05}{sup +0.03} GeV{+-}1.1{center_dot}{delta}M{sub {chi}-}tilde{sub 1}{sup 0} and 183{sub -5}{sup +11} GeV{+-}18{center_dot}{delta}M{sub {chi}-}tilde{sub 1}{sup 0}, respectively, while the true values in SPS1a' are 107.9 GeV and 194.9 GeV, respectively. This corresponds to {delta}M{sub {tau}-}tilde{sub 1}/M{sub {tau}-}tilde{sub 1{approx}}0.1% and {delta}M{sub {tau}-}tilde{sub 2}/M{sub {tau}-}tilde{sub 2{approx}}3% with reasonable assumptions on {delta}M{sub {chi}-}tilde{sub 1}{sup 0}. The cross sections for {tau}-tilde{sub 1} and {tau}-tilde{sub 2} pair production could be obtained with a precision of 3.1% and 4.2%, respectively. Combining the mass and cross section measurement in the {tau}-tilde{sub 2} channel allows to determine the LSP mass with a relative error of 1.7%, assuming a known {tau}-tilde mixing angle. In {tau}-tilde{sub 1{yields}{tau}{chi}}-tilde{sub 1}{sup 0} decays, the {tau} polarization is measured to be 91{+-}9% and 86{+-}5% in the {pi} and {rho} decay channels of the {tau}, respectively. The true value in the simulated SPS1a' sample is 89.6%.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Toward a better understanding of the gamma index: Investigation of parameters with a surface-based distance method

Purpose: The purpose of this work was to clarify the interactions between the parameters used in the {gamma} index with the surface-based distance method, which itself can be viewed as a generalized version of the {gamma} index. The examined parameters included the distance to agreement (DTA)/dose difference (DD) criteria, the percentage used as a passing criterion, and the passing percentage for given DTA/DD criteria. The specific aims of our work were (1) to understand the relationships between the parameters used in the {gamma} index, (2) to determine the detection limit, or the minimum detectable error, of the {gamma} index with a given set of parameters, and (3) to establish a procedure to determine parameters that are consistent with the capacity of an IMRT QA system. Methods: The surface-based distance technique with dose gradient factor was derived, and then the relationship between surface-based distance and {gamma} index was established. The dose gradient factor for plans and measurements of 10 IMRT patients, 10 spine stereotactic radiosurgery (SRS) patients, and 3 Radiological Physics Center (RPC) head and neck phantom were calculated and evaluated. The detection limits of the surface-based distance and {gamma} index methods were examined by introducing known shifts to the 10 IMRT plans. Results: The means of the dose gradient factors were 0.434 mm/% and 0.956 mm/% for the SRS and IMRT plans, respectively. Key quantities (including the mean and 90th and 99th percentiles of the distance distribution) of the surface-based distance distribution between two dose distributions were linearly proportional to the actual shifts. However, the passing percentage of the {gamma} index for a given set of DTA/DD criteria was not associated with the actual shift. For IMRT, using the standard quality assurance criteria of 3 mm/3% DTA/DD and a 90% passing rate, we found that the detection limit of the {gamma} index in terms of global shift was 4.07 mm/4.07 % without noise. Conclusions: Surface-based distance is a direct measure of the difference between two dose distributions and can be used to evaluate or determine parameters for use in calculating the {gamma} index. The dose gradient factor represents the weighting between spatial and dose shift and should be determined before DTA/DD criteria are set. The authors also present a procedure to determine {gamma} index parameters from measurements.

61 RADIATION PROTECTION AND DOSIMETRY↗

An indirect approach to optimize the reaction rates of thermal NO formation for diesel engines

With stringent emission regulations, it has become more important for modern diesel engine manufacturers to accurately predict engine-out nitrogen oxide (NO x ) emissions across a wide range of operating conditions. Thermal NO is the major source of engine-out NO x in modern diesel engines. For thermal NO formation, several earlier studies have recommended the forward and reverse reaction rate coefficients of the rate-limiting reaction (O + N 2 ⇌ NO + N). However, due to deficiencies in sub-models and inadequacies of reduced chemical mechanisms to represent diesel combustion, these recommended values more often than not need to be adjusted in reduced order combustion models to accurately predict engine-out NO x . Hence, in this work a systematic and computationally efficient approach has been proposed to streamline the process of determining the optimum reaction rate coefficients. Here, to develop the optimization approach, four different production diesel engines with different operating conditions in terms of speed, load, and exhaust-gas recirculation have been considered. Numerical simulations have been performed using a detailed zero-dimensional velocity-composition-frequency transported probability density function (0D-VCF-tPDF) model that uses hundreds of notional particles to capture in-cylinder stratification. Four different combinations of hydrocarbon and NO x chemical mechanisms were used to represent chemistry. It was found that for the rate-limiting reaction, the pre-exponent factors (A f1 , A r1 ) and activation energies (E A,f1 , E A,r1 ) of the forward and reverse reaction rates follow a linear band in A f1 - E A,f1 and A r1 - E A,r1 space where predicted engine-out NO x match the measured values closely. By encompassing such bands from different engines and considering constraints on activation energies, a reduced search domain of pre-exponent factors and activation energies was constructed that is expected to be applicable to any diesel engine. Eventually, computationally efficient three-line and one-line search approaches were proposed to determine the optimum values of the pre-exponent factors and activation energies that led to a minimum error between measured and predicted engine-out NO x . Finally, these three-line and one-line NO x optimization approaches were applied to a fifth production diesel engine for which the 0D-VCF-tPDF model showed a very good predictive performance in terms of predicting peak pressure, 50% burn rate, and engine-out NO x when compared to measured and 3D-CFD values.

33 ADVANCED PROPULSION SYSTEMS↗

An a priori evaluation of a principal component and artificial neural network based combustion model in diesel engine conditions

A principal component analysis (PCA) and artificial neural network (ANN) based chemistry tabulation approach is presented. ANNs are used to map the thermochemical state onto a low-dimensional manifold consisting of five control variables that have been identified using PCA. Three canonical configurations are considered to train the PCA-ANN model: a series of homogeneous reactors, a nonpremixed flamelet, and a two-dimensional lifted flame. The performance of the model in predicting the thermochemical manifold of a spatially-developing turbulent jet flame in diesel engine thermochemical conditions is a priori evaluated using direct numerical simulation (DNS) data. The PCA-ANN approach is compared with a conventional tabulation approach (tabulation using ad hoc defined control variables and linear interpolation). The PCA-ANN model provides higher accuracy and requires several orders of magnitude less memory. Here, these observations indicate that the PCA-ANN model is superior for chemistry tabulation, especially for modelling complex chemistries that present multiple combustion modes as observed in diesel combustion. The performance of the PCA-ANN model is then compared to the optimal estimator, i.e. the conditional mean from the DNS. The results indicate that the PCA-ANN model gives high prediction accuracy, comparable to the optimal estimator, especially for major species and the thermophysical properties. Higher errors are observed for the minor species and reaction rate predictions when compared to the optimal estimator. It is shown that the prediction of minor species and reaction rates can be improved by using training data that exhibits a variation of parameters as observed in the turbulent flame. The output of the ANN is analysed to assess mass conservation. It is observed that the ANN incurs a mean absolute error of 0.05% in mass conservation. Furthermore, it is demonstrated that this error can be reduced by modifying the cost function of the ANN to penalise for deviation from mass conservation.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Parameterization Strategies for Intermolecular Potentials for Predicting Trajectory-Based Collision Parameters

The accuracy of separable strategies for constructing full-dimensional potential energy surfaces for collisional energy transfer and collision rate calculations is studied systematically for three alcohols (A = methanol, ethanol, and butanol) and three bath gases (M = Ar, N 2 , and H 2 O). The fitting efficiency (defined as the number of ab initio data required to achieve parametrizations of a desired accuracy) is quantified for both pairwise (Buckingham or "exp6") and nonpairwise (permutationally invariant polynomials, PIPs, of Morse variables) functional forms, for four sampling strategies, and as a function of the complexity and anisotropy of the interaction potential. We find that convergence with respect to the number of sampled ab initio data is largely independent of the choice of functional form but instead varies nearly linearly with the number of adjustable parameters and depends strongly on the sampling strategy. Specifically, the use of biased Sobol quasirandom sampling is similar to 7X more efficient than using unbiased pseudorandom sampling, on average, requiring just similar to 3 computed ab initio energies per adjustable fitting parameter. The pairwise exp6 functional form is shown to provide accurate and transferable parametrizations for M = Ar but is unable to accurately describe alcohol interactions with M = N 2 and H 2 O. The nonpairwise PIP functional form, which is systematically improvable, can produce separable parametrizations with arbitrarily small fitting errors. However, these can suffer from overfitting, which is demonstrated using dynamics calculations of collision parameters for a large number of exp6 and PIP parametrizations. The tests described here validate a robust strategy for automatically generating A + M potential energy surfaces with minimal human intervention, including a quantifiable out of sample metric for judging the accuracy of the fitted surface. We further analyze this set of automatically generated potential energy surfaces to identify areas where more sophisticated fitting strategies may be desired, including pruning of the PIP expansions for large systems and improved sampling strategies more closely coupled with the description of the functional form.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗