betaFeedback

WorksComputer Science

Skeletal Reaction Models for Gasoline Surrogate Combustion

Abstract

Skeletal reaction models are derived for a four-component gasoline surrogate model via an instantaneous local sensitivity analysis technique. The sensitivities of the species mass fractions and the temperature with respect to the reaction rates are estimated by a reduced-order modeling (ROM) methodology. Termed "implicit time-dependent basis CUR (implicit TDB-CUR)," this methodology is based on the CUR matrix decomposition and incorporates implicit time integration for evolving the bases. The estimated sensitivities are subsequently analyzed to develop skeletal reaction models with a fully automated procedure. The 1389-species gasoline surrogate model developed at Lawrence Livermore National Laboratory (LLNL) is selected as the detailed kinetics model. The skeletal reduction procedure is applied to this model in a zero-dimensional constant-pressure reactor over a wide range of initial conditions. The performances of the resulting skeletal models are appraised by comparison against the results via the LLNL detailed model, and also predictions via other skeletal models. Two new skeletal models are developed consisting of 679 and 494 species, respectively. The first is an alternative to an existing model with the same number of species. The predictions with this model reproduces the detailed models vital flame results with less than 1% errors. The errors via the second model are less than 10%.

Introduction

Gasoline fuels consist of hundreds of hydrocarbons. Computational modeling and simulation of such complex chemical mixtures require accounting for numerous species mass fractions. To simplify this, gasoline surrogate mixtures are formulated with fewer hydrocarbons. These surrogates mimic the chemical kinetics and thermodynamic properties of real gasolines [1, 2]. However, surrogate models still involve O(1000)O(1000) species, creating significant computational challenges, especially for multidimensional reacting flow simulations. Therefore, developing skeletal kinetics models for gasoline surrogates by eliminating unimportant species and reactions from the detailed models is essential for feasible simulations.

During the last two decades, a variety of techniques have been developed for skeletal kinetics reduction, by systematical eliminations of unimportant species and reactions while maintaining the overall predictive capability [3, 4, 5]. Local sensitivity analysis (LSA), reaction flux analysis [6, 7], computational singular perturbation (CSP) analysis [8, 9, 10, 11], and directed relation graph (DRG) and its variants [12, 13, 14, 15] have often been utilized. The LSA explores the response of model outputs to small changes in parameters from their nominal values [16]. Global sensitivity analysis considers uncertainty in the kinetic parameters (i.e. collision frequencies and activation energies) and non-linear coupling effects [17, 18, 19, 20, 21, 22, 23, 24, 25]. Instantaneous sensitivity analysis evaluates sensitivities at a specific time in a dynamic system, while time-independent sensitivity analysis measures how values change with parameter variations in a time-averaged sense [26, 27]. Instantaneous LSA-based techniques, which are computationally costly for large kinetics models, include methods such as the principal component analysis (PCA) [28, 29, 30, 31, 32, 33, 34], and species ranking construction [35]. The f-OTD method is specifically designed for low-rank approximations in instantaneous local sensitivity calculations. In Refs. [36, 37, 38] the f-OTD is employed to approximate sensitivities to generate skeletal models for hydrocarbon combustion and nuclear burning. In order to enhance the computational advantages of f-OTD and to overcome its numerical challenges, a new method termed “implicit time-dependent basis CUR (implicit TDB-CUR)” has been developed [39, 40]. The CUR denotes the CUR matrix decomposition, which constructs a low-rank approximation matrix in terms of a subset of its columns (C), a core matrix (U), and a subset of its rows (R) [41, 42]. The implicit TDB-CUR method combines sparse sampling, CUR decomposition, and a cost-efficient implicit time integration algorithm, offering improved efficiency and stability compared to f-OTD. These enhancements are particularly useful for dealing with stiff reaction networks characterized by reactions with orders of magnitude different time scales. Gasoline surrogate models are among these reaction networks.

The Lawrence Livermore National Laboratory (LLNL) gasoline surrogate model [43] contains the chemistry parameters for various hydrocarbon compounds including iso-octane, n-heptane, toluene, and C5-C6 olefins. This model has been validated against a wide range of experimental data, making it broadly applicable for predicting various gasoline surrogate combustion phenomena [43, 44, 45]. However, with 1389 species and 10734 irreversible reactions, the model is computationally prohibitive for practical simulations. Mehl et al. [46] developed a 312-species model using a computer-assisted reduction mechanism, followed by a search algorithm that sequentially assessed the importance of each species. Despite its compact size, the model is not sufficiently accurate (see § 3.2). Lu et al. [47] constructed a reasonably accurate 679-species skeletal model but still left room for further size reduction. Niemeyer et al. [48] derived two reduced models containing 97 and 79 species, each tailored to different operating conditions. Their approach incorporated a directed relation graph with error propagation and sensitivity analysis (DRGEPSA) [15], unimportant reactions … tion elimination, isomer lumping, and analytical quasi-steady-state (QSS) assumption. While isomer lumping and QSS significantly reduce model size, they compromise model versatility by neglecting isomer-specific reactivity differences and assuming rapid equilibration of certain species.

The objective of this work is to develop accurate skeletal models from the LLNL gasoline surrogate model using sensitivities estimated via the implicit TDB-CUR method. This reduction targets a four-component gasoline surrogate (iso-octane, n-heptane, 2-pentene, and toluene), formulated by Mehl et al. [46] and focuses on a lean-to-stoichiometric, low-temperature, homogeneous charge compression ignition (HCCI)-like range of conditions. The composition of the surrogate fuel is given in Table 1. The models are evaluated alongside two previously developed reduced models, and the detailed model to predict the ignition delay, the flame speed, and the flame structure. The efficiency of the implicit TDB-CUR sensitivity approximation is also assessed via comparison against the full-size sensitivity integration.

ComponentsMole Fractions
iso-octane48.70%
n-heptane15.45%
toluene30.60%
2-pentene5.25%

Table 1. Composition of the Surrogate Fuel

Methodology

The reduction procedure consists of the followings: First, the implicit TDB-CUR method with the fourth-order backward differentiation formula (BDF4) [49, 50, 51] is used to compute sensitivities across multiple initial conditions (cases). These sensitivities are then collected and analyzed to produce a ranking of the species and the reactions. Finally, reduced models are constructed based on these rankings.

Sensitivity Analysis

As described in Refs. [36, 37, 52], local linear sensitivity coefficients in a chemical kinetics system is evaluated by infinitesimal perturbations of all reaction rates. The full-size matrix containing all the sensitivity coefficients evolves based on the following matrix differential equation:

dS(t)dt=L(t)S(t)+F(t),(1)\frac{dS(t)}{dt} = L(t)S(t) + F(t), \tag*{(1)}

where S(t)∈Rneq×nrcS(t) \in\mathbb{R}^{n_{\mathrm{eq}} \times n_{\mathrm{rc}}} is the sensitivity matrix, L(t)∈Rneq×neqL(t) \in\mathbb{R}^{n_{\mathrm{eq}} \times n_{\mathrm{eq}}} is the Jacobian matrix, and F(t)∈Rneq×nrcF(t) \in\mathbb{R}^{n_{\mathrm{eq}} \times n_{\mathrm{rc}}} is the forcing term. The number of reactions is noted as nrcn_{\mathrm{rc}}. The number of state variables is neq=nsp+1n_{\mathrm{eq}} = n_{\mathrm{sp}} + 1 where nspn_{\mathrm{sp}} denotes the number of chemical species. Temperature is also solved as a state variable. Equation (1) represents the full-order model (FOM). The implicit TDB-CUR method aims to accurately approximate the sensitivity matrix Sk=S(tk)S^k = S(t^k) at all time steps. The low-rank components are stored in a singular value decomposition (SVD)-like factorized form: Sk=UkΣkYkS^k = U^k\Sigma^kY^k. To compute the low-rank components for the next time step tk+1t^{k+1}, the CUR algorithm is applied, requiring the solution of selected columns and rows of the sensitivity matrix.

In CUR decomposition, the sensitivity matrix S(t)∈Rneq×nrcS(t) \in\mathbb{R}^{n_{eq}\times n_{rc}} is approximated by a subset of its columns and rows, S≈CURS \approx CUR, where C∈Rneq×rC \in\mathbb{R}^{n_{eq}\times r} consists of rr selected columns of SS, R∈Rr×nrcR \in\mathbb{R}^{r\times n_{rc}} consists of rr selected rows of SS, and U∈Rr×rU \in\mathbb{R}^{r\times r} is a low-rank coefficient matrix that captures the interaction between the selected columns and rows. Unlike singular value decompositions, CUR provides a more interpretable decomposition, as the selected columns and rows correspond directly to actual sensitivity information rather than abstract linear combinations. In the implicit TDB-CUR, this decomposition is applied to the sensitivity matrix at each time step, maintaining a low-rank representation without requiring additional evolving equations for the low-rank components. The columns and rows are integrated implicitly using the BDF4 method. The updated sensitivities are stored in the same format: Sk+1=Uk+1Σk+1Yk+1S^{k+1} = U^{k+1}\Sigma^{k+1}Y^{k+1}. To obtain the initial conditions of the low-rank components, the FOM is integrated for several time steps. Then the full-order sensitivity matrix So=S(to)S^o = S(t^o) is factorized by a SVD. This factorization leads to three low-rank components: Uo∈Rneq×rU^o \in\mathbb{R}^{n_{eq}\times r}, Σo∈Rr×r\Sigma^o \in\mathbb{R}^{r\times r}, and Yo∈Rnrc×rY^o \in\mathbb{R}^{n_{rc}\times r}, such that So≈UoΣoYoTS^o \approx U^o\Sigma^oY^{oT}, where rr denotes the number of retained modes (r≪neqr \ll n_{eq}). Then, instead of integrating the FOM, the three low-rank components are updated and saved at each time step. In this way, the computing and memory cost of implicit sensitivities integration is substantially reduced in comparison to that required for the solution of the FOM.

The integration of selected columns and rows involves solving the following equations:

Sk+1(:,Ic)=Sk(:,Ic)+δSk(:,Ic),(2)S^{k+1}(:,I_c) = S^k(:,I_c) + \delta S^k(:,I_c), \tag*{(2)}
Sk+1(Ir,:)=Sk(Ir,:)+δSk(Ir,:),(3)S^{k+1}(I_r,:) = S^k(I_r,:) + \delta S^k(I_r,:), \tag*{(3)}

where δSk(:,Ic)\delta S^k(:,I_c) and δSk(Ir,:)\delta S^k(I_r,:) denote the correction terms for the columns IcI_c and rows IrI_r, respectively. Matlab-style indexing is adopted, where Sk(:,Ic)S^k(:,I_c) denotes a submatrix containing the specified columns, and Sk(Ir,:)S^k(I_r,:) denotes a submatrix containing the specified rows. The GappyPOD+E method [53] is used to select IcI_c and IrI_r to minimize residuals from the low-rank approximation. The column correction term is obtained by solving the linear system:

AkδSk(:,Ic)=bk,(4)A^k\delta S^k(:,I_c) = b^k, \tag*{(4)}

where AkA^k and bkb^k depend on the implicit integrator. For the BDF4 scheme: Ak=(I−1225ΔtL)A^k = (I - \frac{12}{25}\Delta t L) and bk=[4825Sk(:,Ic)−3625Sk−1(:,Ic)+1625Sk−2(:,Ic)−325Sk−3(:,Ic)]−Sik+1(:,Ic)+1225Δt(LkSk(:,Ic)+Fk(:,Ic))b^k = [\frac{48}{25}S^k(:,I_c) - \frac{36}{25}S^{k-1}(:,I_c) + \frac{16}{25}S^{k-2}(:,I_c) - \frac{3}{25}S^{k-3}(:,I_c)] - S_i^{k+1}(:,I_c) + \frac{12}{25}\Delta t(L^k S^k(:,I_c) + F^k(:,I_c)), where II denotes the identity matrix. Since the local sensitivity system is linear, only one linear solve is required. For the FOM, the same formula is used, except that IcI_c includes all of the columns.

Row integration requires a special treatment. In a linear system of equations, the solution for any entry in the unknown vector is coupled to all other entries in that vector. Therefore, integrating the selected rows requires solving the entire linear system. To reduce the computational cost, a low-rank approximation of δSk(Ir,:)\delta S^k(I_r,:) is employed in implicit TDB-CUR. Resulting a reduced-order linear system:

A(r)kδSk(Ir,:)=b(r)k,(5)A_{(r)}^k \delta S^k(I_r,:) = b_{(r)}^k, \tag*{(5)}

where A(r)k=Ak(Ir,Ir)+Ak(Ir,I~r)Uσk(I~r,:)Uσk(Ir,:)†A_{(r)}^k = A^k(I_r,I_r) + A^k(I_r,\widetilde{I}_r)U_{\sigma}^k(\widetilde{I}_r,:)U_{\sigma}^k(I_r,:)^\dagger and b(r)k=[4825Sk(Ir,:)−3625Sk−1(Ir,:)+1625Sk−2(Ir,:)−35Sk−3(Ir,:)]−Sk+1(Ir,:)+1225Δt(Lk(Ir,:)Sk+Fk(Ir,:))b_{(r)}^k = [\frac{48}{25}S^k(I_r,:) - \frac{36}{25}S^{k-1}(I_r,:) + \frac{16}{25}S^{k-2}(I_r,:) - \frac{3}{5}S^{k-3}(I_r,:)] - S^{k+1}(I_r,:) + \frac{12}{25}\Delta t(L^k(I_r,:)S^k + F^k(I_r,:)). I~r\widetilde{I}_r denotes the complement set of IrI_r and the †\dagger sign denotes a pseudo-inverse. The orthonormal basis UσkU_{\sigma}^k is obtained by a SVD operation on a matrix SσkS_{\sigma}^k that is obtained by subtracting previous column solutions from the tk+1t^{k+1} column solution: Sσk=Sk+1(:,Ic)−Sk(:,Ic)−Sk−1(:,Ic)−Sk−2(:,Ic)−Sk−3(:,Ic)S_{\sigma}^k = S^{k+1}(:,I_c) - S^k(:,I_c) - S^{k-1}(:,I_c) - S^{k-2}(:,I_c) - S^{k-3}(:,I_c). Equation (4), with A(r)k∈Rr×rA_{(r)}^k \in\mathbb{R}^{r \times r}, can be solved efficiently to find δSk(Ir,:)\delta S^k(I_r,:).

The low-rank components of sensitivity matrix at tk+1t^{k+1} is computed as follows:

Sk+1(:,Ic)=UΣYT,(6)S^{k+1}(:,I_c) = U\Sigma Y^T, \tag*{(6)}
V=[U(Ir,:)′Sk+1(Ir,:)]T,(7)V = [U(I_r,:)'S^{k+1}(I_r,:)]^T, \tag*{(7)}
V=Yk+1Σk+1RU,(8)V = Y^{k+1}\Sigma^{k+1}R_U, \tag*{(8)}
Uk+1=URU,(9)U^{k+1} = UR_U, \tag*{(9)}

where UU is computed via a SVD on column solutions. Σk+1\Sigma^{k+1} and Yk+1Y^{k+1} are obtained by a SVD on the matrix VV. Finally, the remaining sensitivity component Uk+1U^{k+1} is obtained by rotating the basis UU.

Reactions and Species Ranking

After collecting low-rank sensitivity components from multiple cases, the reactions and the species can be ranked. The right singular vectors Yk=Y(tk)Y^k = Y(t^k), which reflect the reaction importance, is weighted by the singular values Σk\Sigma^k:

W(j)k=∑n∣Ynk∣Σnk/∑nΣnk,(10)W_{(j)}^k = \sum_n \lvert Y_n^k\rvert\Sigma_n^k \Big/ \sum_n \Sigma_n^k, \tag*{(10)}

where the index nn denotes the nnth right singular vector or singular value. Each element of W(j)k∈RnrcW_{(j)}^k \in\mathbb{R}^{n_{rc}}, is positive and is associated with a certain reaction. The larger the associated WW value, the more important the reaction is. The parameter χ∈Rnrc\chi\in\mathbb{R}^{n_{rc}} is defined as the highest value associated with WW across all resolved time-steps and cases. The elements of the χ\chi vector are sorted in descending order to find the indices of the most important reactions in the detailed model. The species are ranked based on their first presence in the sorted reactions, i.e. a species that first appears in a higher-ranked reaction is considered more important

Initial ConditionsTime StepsFOMImplicit TDB-CUR
Case 1T800, P10, ϕ\phi1.064666320.6 sec3587.3 sec
Case 2T750, P60, ϕ\phi1.068266905.2 sec3874.6 sec
Case 3T1200, P60, ϕ\phi0.645443633.8 sec2080.0 sec
Case 4T1100, P10, ϕ\phi0.647663891.5 sec2210.6 sec
Case 5T1000, P60, ϕ\phi0.649484538.6 sec2575.6 sec
Case 6T800, P10, ϕ\phi0.665666239.1 sec3593.2 sec
Case 7T750, P10, ϕ\phi0.668206629.3 sec3842.9 sec
Case 8T750, P60, ϕ\phi0.667646774.1 sec3847.3 sec
Case 9T800, P60, ϕ\phi0.244403031.3 sec1745.5 sec
Case 10T700, P20, ϕ\phi0.248523209.0 sec1838.3 sec
Case 11T800, P20, ϕ\phi0.242443435.2 sec1956.3 sec

caption: tbl:2: Clock Times for Computing Sensitivities

:::

than one that first participates in a lower-ranked reaction. This process results in a reaction and species ranking based on the χ\chi vector. The reduced model is constructed by removing the less important species from the species ranking while retaining all reactions associated with the remaining species.

The skeletal reduction scheme as shown here is based on instantaneous local sensitivities. Unlike PCA-based sensitivity methods [29], where full sensitivities are computed first and then reduced, the implicit TDB-CUR approach solves for the sensitivities directly in their reduced form. This leads to computational savings in floating-point operations (FLOPs) and substantial reductions in memory usage, while also eliminating the need to store full-size sensitivity data.

Skeletal Reduction of Gasoline Surrogate

The implicit TDB-CUR method with the BDF4 integrator [51] is employed for low-rank approximation of gasoline surrogate model sensitivities. A total of 11 cases are considered, with initial conditions summarized in Table 2 taken from Refs. [46, 48]. These cases cover the ranges of 750 KK to 1200 KK for the temperature, 10 atmatm to 60 atmatm for the pressure, and 0.2 to 1.0 for the equivalence ratio. The CUR rank is set as r=7r = 7. To benchmark the efficiency and the accuracy of implicit TDB-CUR, the FOM is also solved for all the cases by the BDF4. Following the FOM solution, an SVD of the sensitivity matrix is performed at each time step. The data from these SVD is referred to as the instantaneous SVD (iSVD). The seven largest singular values from the iSVD are compared with those obtained using the implicit TDB-CUR. The sensitivities estimated by the implicit TDB-CUR are further analyzed to generate species and reaction rankings. The performance of the reduced models is evaluated by comparison against the results of the detailed model. Two existing reduced models are also included for comparison: chem312 [46] and chem679 [47].

TDB-CUR Evaluation

Figure 1 provides a comparison of the singular values of the sensitivity matrix obtained by implicit TDB-CUR against those via iSVD for Case 1. It is shown that there is one dominant mode for which the singular value is multiple orders of magnitude larger than the others. The implicit TDB-CUR is able to accurately capture the first three singular values. The smaller modes do not align with the FOM results, but have no

Singular values

Figure 1. Singular values in case 1 (ϕ=1.0\phi= 1.0, T=800K, P=10 atm).

Modelnspn_{sp}nrcn_{rc}
Detailed138910734
rd6796795731
rd4944944154

caption: Chemistry Models

:::

significant effects on the predicted results, since the right singular values are weighted by the singular values.

As for the efficiency of the implicit TDB-CUR, the clock time of ROM simulation is compared against FOM in Table 2. In all the cases, the ROM requires less than 60% time of the FOM.

Skeletal Model Validation

In Figure 2, the top 70 most important species and reactions of the LLNL gasoline surrogate model are ranked based on their importance. The higher the χ\chi values, the higher the importance of a species/reaction. The Appendix provide the complete ranking of the species and rankings. In the reactions ranking, a toluene oxidization reaction (Reaction 9859 : C6H5CH3+O2→C6H5CH2J+HO2C_6H_5CH_3 + O_2 \to C_6H_5CH_2J + HO_2) is identified as the most important. The species involved in this reaction are identified as the most important species, as indicated by the species ranking. The second most important is an iso-octane pyrolysis reaction (Reaction 5776 : IC8H18→CH3+YC7H15IC_8H_{18} \to CH_3 + Y C_7H_{15}). This is consistent with the findings in Ref. [54]. Based on the species ranking two reduced reaction models are suggested. The first contains the top 679 most important species (rd679), the same number as that in the chem679 [47]. This model yields less than %1 error. The second model with the top 494 species (rd494) guarantees 10% maximum relative error of all comparative assessments with the minimum number of species. The numbers of reactions for these reduced models are listed in Table 3. The reaction size refers to the number of irreversible reactions.

Image not extracted

Figure 2. Importance ranking of the species and reactions. Reaction ranking is labeled by the reaction indices associated with the supplemental Cantera file

In Figure 3, the ignition delays are shown for several equivalence ratios and pressure values. The relative errors in predicting the delays are also provided and indicate that rd679 and chem679 provide the best overall accuracy. The relative errors associated with these two models are of order 1%. For the rd494 model, the relative errors are less than 8% in most cases. With ϕ=0.6\phi= 0.6 & 1.0, the rd494 models show larger errors when 1000/T0=1.1∼1.4 K−11000/T_0 = 1.1 \sim1.4\ \mathrm{K}^{-1}, which correspond to regions with lower gradients. This behavior is known as the “negative temperature coefficients” (NTC) [55]. The shape of the rd494 relative error curves indicates that the NTC may increase the difficulty of producing accurate reduced models using sensitivity coefficients. The chem312 model [46] provides errors up to 30%, but yields accurate predictions when P=10 atmP = 10\ \mathrm{atm} and NTC behavior occurs.

Ignition delays and relative errors for the reduced models

Figure 3. Left: Ignition delay test of the reduced models. Right: Relative errors ϵid=∣τreduced−τdetailed∣/τdetailed\epsilon_{id} = \lvert\tau_{reduced} - \tau_{detailed}\rvert/\tau_{detailed}.

Simulations are also conducted to determine the laminar flame speed with T0=300 kT_0 = 300\ \mathrm{k}. Figure 4 shows that all the relative errors are within 6%. The rd494 model provides less than 3% error in most cases, while rd679 yields less than 2% error for all cases. The 1D flame structures are determined for several pressure values. In Figure 5, the distribution of temperatures and mole fractions of O2O_2, CO2CO_2, and IC8H18IC_8H_{18} corroborate that the rd494 model perfectly replicates the 1D flame structure.

Flame Speeds and Relative Errors

Figure 4. Left: Laminar flame speed test (T0=300 kT_0 = 300\ k). Right: Relative errors ϵfs=∣SL,reduced0−SL,detailed0∣/SL,detailed0\epsilon_{fs} = \lvert S^{0}_{L,\mathrm{reduced}} - S^{0}_{L,\mathrm{detailed}} \rvert/S^{0}_{L,\mathrm{detailed}}.

1D flame structure test of the rd494 model

Figure 5. 1D flame structure test of the rd494 model. Left: The distribution of flame temperatures. Right: Mole fractions of key species.

Conclusions

The implicit time-dependent basis CUR (implicit TDB-CUR) method is employed for low-rank approximation of linear sensitivities in the gasoline surrogate combustion based on the Lawrence Livermore National Laboratory (LLNL) kinetics model. It is shown that this method evaluates sensitivities with computational costs significantly lower than that required to solve the full-order model (FOM). The computed sensitivities are then analyzed for skeletal reduction of the LLNL surrogate fuel model. In doing so, the species are ranked based on their sensitivities to the reaction rates. The rd679 skeletal model as developed here shows the same accuracy as the chem679 model developed in Ref. [47]. The rd494 with a lower number of species is also generated. The accuracy and efficiency of the model are assessed by comparing its predictions for the ignition

delay, the flame speed, and 11-D flame structure. The rd494 model, which contains about 35% of the total species suggested by the LLNL model, while yielding better than 90% accuracy, is a reliable and affordable alternative to the detailed model.

This work demonstrates that the implicit TDB-CUR method accurately approximates the sensitivities and generates skeletal reaction models in an automated and efficient manner. The ability to perform such a reduction automatically for large and complex systems eliminates the need for user expertise in chemical kinetics networks, marking a significant advancement. Although the reduction scheme is applied to a specific gasoline surrogate mixture, it is broadly applicable to the reduction of other multi-component fuels. Moreover, it is well-suited for reducing other reaction networks, such as nuclear [56] and biochemical systems [57].

Acknowledgments

This work is co-authored by an employee of Triad National Security, LLC which operates Los Alamos National Laboratory under Contract No.89233218CNA000001 with the U.S. Department of Energy/National Nuclear Security Administration. The work at Pitt is sponsored by the NSF under Grant CBET-2042918 and Grant CBET-2152803. Computational resources are provided by the Center for Research Computing (CRC) at Pitt.

Appendix

All of the reduced models can be accessed via this link.

References

  1. [1]X. Zhen, Y. Wang, D. Liu, An Overview of the Chemical Reaction Mechanisms for Gasoline Surrogate Fuels, Appl. Therm. Eng. 124 (2017) 1257–1268.DOI
  2. [2]S. M. Sarathy, A. Farooq, G. T. Kalghatgi, Recent Progress in Gasoline Surrogate Fuels, Prog. Energy Combust. Sci. 65 (2018) 67–108.DOI
  3. [3]C. K. Law, Combustion Physics, Cambridge University Press, 2010.
  4. [4]L. P. de Oliveira, D. Hudebine, D. Guillaume, J. J. Verstraete, A Review of Kinetic Modeling Method-ologies for Complex Processes, Oil Gas Sci Technol–Revue d’IFP Energies nouvelles 71 (3) (2016) 45.
  5. [5]D. Goussis, U. Maas, Model Reduction for Combustion Chemistry, in: T. Echekki, E. Mastorakos (Eds.), Turbulent Combustion Modeling, Springer, 2011, pp. 193–220.DOI
  6. [6]W. Sun, Z. Chen, X. Gou, Y. Ju, A Path Flux Analysis Method for the Reduction of Detailed Chemical Kinetic Mechanisms, Combust. Flame 157 (7) (2010) 1298–1307.DOI
  7. [7]W. Wang, X. Gou, An Improved Path Flux Analysis with Multi Generations Method for Mechanism Reduction, Combust. Theory Model. 20 (2) (2016) 203–220.DOI
  8. [8]S. H. Lam, D. A. Goussis, Understanding Complex Chemical Kinetics with Computational Singular Perturbation, Proc. Combust. Inst. 22 (1) (1989) 931–941.DOI
  9. [9]S. H. Lam, Using CSP to Understand Complex Chemical Kinetics, Combust. Sci. Technol. 89 (5-6) (1993) 375–404.DOI
  10. [10]S. H. Lam, D. A. Goussis, The CSP Method for Simplifying Kinetics, Int. J. Chem. Kinet. 26 (4) (1994) 461–486.DOI
  11. [11]T. Lu, Y. Ju, C. K. Law, Complex CSP for Chemistry Reduction and Analysis, Combust. Flame 126 (1-2) (2001) 1445–1455.
  12. [12]T. Lu, C. K. Law, A Directed Relation Graph Method for Mechanism Reduction, Proc. Combust. Inst. 30 (1) (2005) 1333–1341.
  13. [13]X. Zheng, T. Lu, C. K. Law, Experimental Counterflow Ignition Temperatures and Reaction Mechanisms of 1, 3-Butadiene, Proc. Combust. Inst. 31 (1) (2007) 367–375.DOI
  14. [14]T. Lu, C. Law, Toward Accommodating Realistic Fuel Chemistry in Large-Scale Computations, Prog. Energy Combust. Sci. 35 (2) (2009) 192–215.DOI
  15. [15]K. Niemeyer, C. Sung, M. Raju, Skeletal Mechanism Generation for Surrogate Fuels using Directed Relation Graph with Error Propagation and Sensitivity Analysis, Combust. Flame 157 (9) (2010) 1760–1770.
  16. [16]S. Li, B. Yang, F. Qi, Accelerate Global Sensitivity Analysis using Artificial Neural Network Algorithm: Case Studies for Combustion Kinetic Model, Combust. Flame 168 (2016) 53–64.
  17. [17]T. Turányi, Sensitivity Analysis of Complex Kinetic Systems. Tools and Applications, J. Math. Chem. 5 (3) (1990) 203–248.DOI
  18. [18]I. Sobol’, On Sensitivity Estimation for Nonlinear Mathematical Models, Matematicheskoe Modelirovanie 2 (1) (1990) 112–118.
  19. [19]T. Homma, A. Saltelli, Importance Measures in Global Sensitivity Analysis of Nonlinear Models, Reliab. Eng. Syst. Saf. 52 (1) (1996) 1–17.DOI
  20. [20]H. Rabitz, Ö. Alış, General Foundations of High-Dimensional Model Representations, J. Math. Chem. 25 (2-3) (1999) 197–233.DOI
  21. [21]I. Sobol’, Global Sensitivity Indices for Nonlinear Mathematical Models and Their Monte Carlo Estimates, Math. Comput. Simul. 55 (1-3) (2001) 271–280.DOI
  22. [22]G. Li, S. Wang, H. Rabitz, Practical Approaches to Construct RS-HDMR Component Functions, J. Phys. Chem. A 106 (37) (2002) 8721–8733.DOI
  23. [23]A. Saltelli, S. Tarantola, F. Campolongo, M. Ratto, Sensitivity Analysis in Practice: A Guide to Assessing Scientific Models, John Wiley & Sons, 2004.
  24. [24]D. A. Sheen, X. You, H. Wang, T. Løvås, Spectral Uncertainty Quantification, Propagation and Optimization of a Detailed Kinetic Model for Ethylene Combustion, Proc. Combust. Inst. 32 (1) (2009) 535–542.DOI
  25. [25]G. Esposito, B. Sarnacki, H. Chelliah, Uncertainty Propagation of Chemical Kinetics Parameters and Binary Diffusion Coefficients in Predicting Extinction Limits of Hydrogen/Oxygen/Nitrogen Non-premixed Flames, Combust. Theory Model. 16 (6) (2012) 1029–1052.
  26. [26]P. Lakrisenko, D. Pathirana, D. Weindl, J. Hasenauer, Benchmarking Methods for Computing Local Sensitivities in Ordinary Differential Equation Models at Dynamic and Steady States, PLOS ONE 19 (10) (2024) e0312148.DOI
  27. [27]P. Lakrisenko, P. Stapor, S. Grein, Ł. Paszkowski, D. Pathirana, F. Fröhlich, G. T. Lines, D. Weindl, J. Hasenauer, Efficient Computation of Adjoint Sensitivities at Steady-state in ODE Models of Biochemical Reaction Networks, PLOS Comput. Biol. 19 (1) (2023) e1010783.DOI
  28. [28]N. Brown, G. Li, M. Koszyкowski, Mechanism Reduction via Principal Component Analysis, Int. J. Chem. Kinet. 29 (6) (1997) 393–414.doi.org/10.1002/(sici)1097-4601(1997)29:6<393::aid-kin1>3.0.co;2-p
  29. [29]G. Esposito, H. Chelliah, Skeletal Reaction Models based on Principal Component Analysis: Application to Ethylene–Air Ignition, Propagation, and Extinction Phenomena, Combust. Flame 158 (3) (2011) 477–489.
  30. [30]A. Parente, J. Sutherland, B. Dally, L. Tognot​ti, P. Smith, Investigation of the MI​LD Combustion Regime via Principal Component Analysis, Proc. Combust. Inst. 33 (2) (2011) 3333–3341.DOI
  31. [31]A. Parente, J. Sutherland, Principal Component Analysis of Turbulent Combustion Data: Data Pre-processing and Manifold Sensitivity, Combust. Flame 160 (2) (2013) 340–350.DOI
  32. [32]H. Mirgolbabaei, T. Echekki, Nonlinear Reduction of Combustion Composition Space with Kernel Principal Component Analysis, Combust. Flame 161 (1) (2014) 118–126.DOI
  33. [33]A. Coussement, B. Isaac, O. Gicquel, A. Parente, Assessment of Different Chemistry Reduction Methods based on Principal Component Analysis: Comparison of the MG-PCA and Score-PCA Approaches, Combust. Flame 168 (2016) 83–97.DOI
  34. [34]M. Malik, B. Isaac, A. Coussement, P. Smith, A. Parente, Principal Component Analysis Coupled with Nonlinear Regression for Chemistry Reduction, Combust. Flame 187 (2018) 30–41.DOI
  35. [35]A. Stagni, A. Frassoldati, A. Cuoci, T. Faravelli, E. Ranzi, Skeletal Mechanism Reduction Through Species-Targeted Sensitivity Analysis, Combust. Flame 163 (2016) 382–393.DOI
  36. [36]A. Nouri, H. Babaee, P. Givi, H. Chelliah, D. Livescu, Skeletal Model Reduction with Forced Optimally Time Dependent Modes, Combust. Flame 235 (2022) 111684.arxiv.org/abs/2103.09101
  37. [37]Y. Liu, H. Babaee, P. Givi, H. Chelliah, D. Livescu, A. Nouri, Skeletal Reaction Models for Methane Combustion, Fuel 357 (2024) 129581.arxiv.org/abs/2201.11038
  38. [38]A. Nouri, Y. Liu, P. Givi, H. Babaee, D. Livescu, Skeletal Kinetics Reduction for Astrophysical Reaction Networks, Astrophys. J. Suppl. S. 272 (2) (2024) 34.arxiv.org/abs/2404.17447
  39. [39]M. Donello, G. Palkar, M. Naderi, D. Del Rey Fernández, H. Babaee, Oblique Projection for Scalable Rank-adaptive Reduced-order Modelling of Nonlinear Stochastic Partial Differential Equations with Time-dependent Bases, Proc. R. Soc. A 479 (2278) (2023) 20230320.DOI
  40. [40]M. H. Naderi, S. Akhavan, H. Babaee, A Cross Algorithm for Implicit Time Integration of Random Partial Differential Equations on Low-rank Matrix Manifolds, Proc. R. Soc. A 481 (2309) (2025) 20240658.arxiv.org/abs/2408.16591
  41. [41]S. A. Goreinov, E. E. Tyrtyshnikov, N. L. Zamarashkin, A Theory of Pseudoskeleton Approximations, Linear Algebra Its Appl. 261 (1-3) (1997) 1–21.DOI
  42. [42]M. W. Mahoney, P. Drineas, CUR Matrix Decompositions for Improved Data Analysis, Proc. Natl. Acad. Sci. U.S.A. 106 (3) (2009) 697–702.DOI
  43. [43]M. Mehl, W. J. Pitz, C. K. Westbrook, H. J. Curran, Kinetic Modeling of Gasoline Surrogate Components and Mixtures Under Engine Conditions, Proc. Combust. Inst. 33 (1) (2011) 193–200.DOI
  44. [44]G. Kukkadapu, K. Kumar, C.-J. Sung, M. Mehl, W. J. Pitz, Experimental and Surrogate Modeling Study of Gasoline Ignition in A Rapid Compression Machine, Combust. Flame 159 (10) (2012) 3066–3078.DOI
  45. [45]G. Kukkadapu, K. Kumar, C.-J. Sung, M. Mehl, W. J. Pitz, Autoignition of Gasoline and Its Surrogates in A Rapid Compression Machine, Proc. Combust. Inst. 34 (1) (2013) 345–352.DOI
  46. [46]M. Mehl, J.-Y. Chen, W. J. Pitz, S. M. Sarathy, C. K. Westbrook, An Approach for Formulating Surrogates for Gasoline With Application Toward a Reduced Surrogate Mechanism for CFD Engine Modeling, Energy & Fuels 25 (11) (2011) 5215–5223.DOI
  47. [47]T. Lu, C. K. Law, Toward Accommodating Realistic Fuel Chemistry in Large-scale Computations, Prog. Energy Combust. Sci. 35 (2) (2009) 192–215.DOI
  48. [48]K. E. Niemeyer, C.-J. Sung, Reduced Chemistry for A Gasoline Surrogate Valid at Engine-relevant Conditions, Energy & Fuels 29 (2) (2015) 1172–1185.DOI
  49. [49]C. F. Curtiss, J. O. Hirschfelder, Integration of Stiff Equations, Proc. Natl. Acad. Sci. U.S.A. 38 (3) (1952) 235–243.pubmed.ncbi.nlm.nih.gov/16589085
  50. [50]C. W. Gear, The Numerical Integration of Ordinary Differential Equations, Math. Comput. 21 (98) (1967) 146–156.DOI
  51. [51]E. Süli, D. F. Mayers, An Introduction to Numerical Analysis, Cambridge university press, 2003.
  52. [52]I. Zsely, J. Zador, T. Turanyi, Similarity of Sensitivity Functions of Reaction Kinetic Models, J. Phys. Chem. A 107 (13) (2003) 2216–2238.DOI
  53. [53]B. Peherstorfer, Z. Drmac, S. Gugercin, Stability of Discrete Empirical Interpolation and Gappy Proper Orthogonal Decomposition with Randomized and Deterministic Sampling Points, SIAM J Sci Comput 42 (5) (2020) A2837–A2864.arxiv.org/abs/1808.10473
  54. [54]G. Ghiasi, I. Ahmed, N. Swaminathan, Gasoline Flame Behavior at Elevated Temperature and Pressure, Fuel 238 (2019) 248–256.DOI
  55. [55]C. K. Law, P. Zhao, NTC-affected Ignition in Nonpremixed Counterflow, Combust. Flame 159 (3) (2012) 1044–1054.DOI
  56. [56]A. G. Nouri, P. Givi, D. Livescu, Modeling And Simulation of Turbulent Nuclear Flames in Type Ia Supernovae, Prog. Aerosp. Sci. 108 (2019) 156–179.DOI
  57. [57]W. W. Chen, M. Niepel, P. K. Sorger, Classic and Contemporary Approaches to Modeling Biochemical Reactions, Genes Dev. 24 (17) (2010) 1861–1875.DOI

Paper details

vlnc:W_wbnexh8q
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Computer Science

Not recorded for this paper

View statement
Looking for related papers…
Contents