George Papadakis
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
George Papadakis has a total of 89 co-authors. Here are the top co-authors by number of shared publications:
- Stelios Rigopoulos
Department of Mechanical Engineering, Imperial College London, London SW7 2AZ, United Kingdom
8 shared publications - James Ferris Whidborne
Cranfield University
6 shared publications - John Christos Vassilicos
Arts et Metiers Institute of Technology
4 shared publications - Kyriakos Dimitrios Kantarakias
Department of Aeronautics, Imperial College London, London SW7 2AZ, United Kingdom
4 shared publications - Michael Yianneskis
Experimental and Computational Laboratory for the Analysis of Turbulence, Division of Engineering, King's College London, London, UK
4 shared publications S.L. Yeoh
4 shared publications
Latest Publications
- Open Access
- 23 July 2026
- Published by Cambridge University Press (CUP) in Data-Centric Engineering
- Vol. 7, e32
Robustness of data assimilation algorithms for reconstructing and forecasting the flow over a surface-mounted prism from sparse data
Three data-driven assimilation algorithms are compared in the flow around a surface mounted prism at $ \mathit{\operatorname{Re}}=1000 $ re equals 1000 . All algorithms employ proper orthogonal decomposition (POD) for dimensionality reduction and synthesize linear estimators (Kalman filters). Two estimators are new and exploit past historical patterns of the training data. They are constructed from the singular vectors of the time-delayed Hankel matrix assembled from the signals of the POD coefficients. The two estimators differ on the forcing term of the linear system and its statistics. They can reconstruct (and also forecast) the instantaneous velocity field from sparse data. The third algorithm is based on the n4sid system identification method. The performance of the algorithms is compared with respect to the number of sensor points, model order, computational time for synthesis, and accuracy at reproducing the statistics of the flow (Reynolds stresses) at design and off-design conditions. For all criteria, it is found that the two new estimators were more robust and accurate compared to n4sid . Between the two, the estimator with random noise forcing was slightly more accurate. The future evolution of the flow was also accurately forecast over long time horizons using only one sensor. Application of the model trained at the reference Reynolds number to an off-design condition (at the nearby $ \mathit{\operatorname{Re}}=800 $ re equals 800 ) showed that very satisfactory results can be obtained with only very few sensors. The new estimators are scalable, easy to construct, and physically interpretable and can provide flow reconstruction (and, if needed, forecasting over a rolling time window) in almost real time.- 23 July 2026
- Published by Cambridge University Press (CUP) in Data-Centric Engineering
- Vol. 7, e32
Open AccessRobustness of data assimilation algorithms for reconstructing and forecasting the flow over a surface-mounted prism from sparse data
Three data-driven assimilation algorithms are compared in the flow around a surface mounted prism at $ \mathit{\operatorname{Re}}=1000 $ re equals 1000 . All algorithms employ proper orthogonal decomposition (POD) for dimensionality reduction and synthesize linear estimators (Kalman filters). Two estimators are new and exploit past historical patterns of the training data. They are constructed from the singular vectors of the time-delayed Hankel matrix assembled from the signals of the POD coefficients. The two estimators differ on the forcing term of the linear system and its statistics. They can reconstruct (and also forecast) the instantaneous velocity field from sparse data. The third algorithm is based on the n4sid system identification method. The performance of the algorithms is compared with respect to the number of sensor points, model order, computational time for synthesis, and accuracy at reproducing the statistics of the flow (Reynolds stresses) at design and off-design conditions. For all criteria, it is found that the two new estimators were more robust and accurate compared to n4sid . Between the two, the estimator with random noise forcing was slightly more accurate. The future evolution of the flow was also accurately forecast over long time horizons using only one sensor. Application of the model trained at the reference Reynolds number to an off-design condition (at the nearby $ \mathit{\operatorname{Re}}=800 $ re equals 800 ) showed that very satisfactory results can be obtained with only very few sensors. The new estimators are scalable, easy to construct, and physically interpretable and can provide flow reconstruction (and, if needed, forecasting over a rolling time window) in almost real time. - Open Access
- 4 May 2026
- Published by Cambridge University Press (CUP) in Journal of Fluid Mechanics
- Vol. 1034
Lyapunov stability analysis of the chaotic flow past two square cylinders
We investigate the stability of the flow past two side-by-side square cylinders (at Reynolds number 200 and gap ratio 1) using tools from dynamical systems theory. The flow is highly irregular due to the complex interaction between the flapping jet emanating from the gap and the vortices shed in the wake. We first perform spectral proper orthogonal decomposition (SPOD) to understand the flow characteristics. We then conduct Lyapunov stability analysis by linearising the Navier–Stokes equations around the irregular base flow and find that it has two positive Lyapunov exponents. The covariant Lyapunov vectors (CLVs) are also computed. Contours of the time-averaged CLVs reveal that the footprint of the leading CLV is in the near-wake, whereas the other CLVs peak further downstream, indicating distinct regions of instability. SPOD of the two unstable CLVs is then employed to extract the dominant coherent structures and oscillation frequencies in the tangent space. For the leading CLV, the two dominant frequencies match closely with the prevalent frequencies in the drag coefficient spectrum and correspond to instabilities due to vortex shedding and jet-flapping. The second unstable CLV captures the subharmonic instability of the shedding frequency. Global linear stability analysis (GLSA) of the time-averaged flow identifies a neutral eigenmode that resembles the leading SPOD mode of the first CLV, with a very similar structure and frequency. However, while GLSA predicts neutrality, Lyapunov analysis reveals that this direction is unstable, exposing the inherent limitations of the GLSA when applied to chaotic flows.
- 4 May 2026
- Published by Cambridge University Press (CUP) in Journal of Fluid Mechanics
- Vol. 1034
Open AccessLyapunov stability analysis of the chaotic flow past two square cylinders
We investigate the stability of the flow past two side-by-side square cylinders (at Reynolds number 200 and gap ratio 1) using tools from dynamical systems theory. The flow is highly irregular due to the complex interaction between the flapping jet emanating from the gap and the vortices shed in the wake. We first perform spectral proper orthogonal decomposition (SPOD) to understand the flow characteristics. We then conduct Lyapunov stability analysis by linearising the Navier–Stokes equations around the irregular base flow and find that it has two positive Lyapunov exponents. The covariant Lyapunov vectors (CLVs) are also computed. Contours of the time-averaged CLVs reveal that the footprint of the leading CLV is in the near-wake, whereas the other CLVs peak further downstream, indicating distinct regions of instability. SPOD of the two unstable CLVs is then employed to extract the dominant coherent structures and oscillation frequencies in the tangent space. For the leading CLV, the two dominant frequencies match closely with the prevalent frequencies in the drag coefficient spectrum and correspond to instabilities due to vortex shedding and jet-flapping. The second unstable CLV captures the subharmonic instability of the shedding frequency. Global linear stability analysis (GLSA) of the time-averaged flow identifies a neutral eigenmode that resembles the leading SPOD mode of the first CLV, with a very similar structure and frequency. However, while GLSA predicts neutrality, Lyapunov analysis reveals that this direction is unstable, exposing the inherent limitations of the GLSA when applied to chaotic flows.
- Open Access
- 2 March 2026
- Published by Cambridge University Press (CUP) in Journal of Fluid Mechanics
- Vol. 1030, A21
Forecasting the evolution of three-dimensional turbulent recirculating flows from sparse sensor data
A data-driven algorithm is proposed that employs sparse data from velocity and/or scalar sensors to forecast the future evolution of three-dimensional turbulent flows. The algorithm combines time-delayed embedding together with Koopman theory and linear optimal estimation theory. It consists of three steps: dimensionality reduction, currently with proper orthogonal decomposition (POD); construction of a linear dynamical system for current and future POD coefficients; and system closure using sparse sensor measurements. In essence, the algorithm establishes a mapping from current sparse data to the future state of the dominant structures of the flow over a specified time window. The method is scalable (i.e. applicable to very large systems), physically interpretable and provides sequential forecasting on a sliding time window of prespecified length. It is applied to the turbulent recirculating flow over a surface-mounted cube (with more than $10^8$ degrees of freedom) and is able to forecast accurately the future evolution of the most dominant structures over a time window at least two orders of magnitude larger that the (estimated) Lyapunov time scale of the flow. Most importantly, increasing the size of the forecasting window only slightly reduces the accuracy of the estimated future states.- 2 March 2026
- Published by Cambridge University Press (CUP) in Journal of Fluid Mechanics
- Vol. 1030, A21
Open AccessForecasting the evolution of three-dimensional turbulent recirculating flows from sparse sensor data
A data-driven algorithm is proposed that employs sparse data from velocity and/or scalar sensors to forecast the future evolution of three-dimensional turbulent flows. The algorithm combines time-delayed embedding together with Koopman theory and linear optimal estimation theory. It consists of three steps: dimensionality reduction, currently with proper orthogonal decomposition (POD); construction of a linear dynamical system for current and future POD coefficients; and system closure using sparse sensor measurements. In essence, the algorithm establishes a mapping from current sparse data to the future state of the dominant structures of the flow over a specified time window. The method is scalable (i.e. applicable to very large systems), physically interpretable and provides sequential forecasting on a sliding time window of prespecified length. It is applied to the turbulent recirculating flow over a surface-mounted cube (with more than $10^8$ degrees of freedom) and is able to forecast accurately the future evolution of the most dominant structures over a time window at least two orders of magnitude larger that the (estimated) Lyapunov time scale of the flow. Most importantly, increasing the size of the forecasting window only slightly reduces the accuracy of the estimated future states. - Open Access
Lyapunov stability analysis of the chaotic flow past two square cylinders
We investigate the stability of the flow past two side-by-side square cylinders (at Reynolds number 200 and gap ratio 1) using tools from dynamical systems theory. The flow is highly irregular due to the complex interaction between the flapping jet emanating from the gap and the vortices shed in the wake. We first perform Spectral Proper Orthogonal Decomposition (SPOD) to understand the flow characteristics. We then conduct Lyapunov stability analysis by linearizing the Navier-Stokes equations around the irregular base flow and find that it has two positive Lyapunov exponents. The Covariant Lyapunov Vectors (CLVs) are also computed. Contours of the time-averaged CLVs reveal that the footprint of the leading CLV is in the near-wake, whereas the other CLVs peak further downstream, indicating distinct regions of instability. SPOD of the two unstable CLVs is then employed to extract the dominant coherent structures and oscillation frequencies in the tangent space. For the leading CLV, the two dominant frequencies match closely with the prevalent frequencies in the drag coefficient spectrum, and correspond to instabilities due to vortex shedding and jet-flapping. The second unstable CLV captures the subharmonic instability of the shedding frequency. Global linear stability analysis (GLSA) of the time-averaged flow identifies a neutral eigenmode that resembles the leading SPOD mode of the first CLV, with a very similar structure and frequency. However, while GLSA predicts neutrality, Lyapunov analysis reveals that this direction is unstable, exposing the inherent limitations of the GLSA when applied to chaotic flows.
Open AccessLyapunov stability analysis of the chaotic flow past two square cylinders
We investigate the stability of the flow past two side-by-side square cylinders (at Reynolds number 200 and gap ratio 1) using tools from dynamical systems theory. The flow is highly irregular due to the complex interaction between the flapping jet emanating from the gap and the vortices shed in the wake. We first perform Spectral Proper Orthogonal Decomposition (SPOD) to understand the flow characteristics. We then conduct Lyapunov stability analysis by linearizing the Navier-Stokes equations around the irregular base flow and find that it has two positive Lyapunov exponents. The Covariant Lyapunov Vectors (CLVs) are also computed. Contours of the time-averaged CLVs reveal that the footprint of the leading CLV is in the near-wake, whereas the other CLVs peak further downstream, indicating distinct regions of instability. SPOD of the two unstable CLVs is then employed to extract the dominant coherent structures and oscillation frequencies in the tangent space. For the leading CLV, the two dominant frequencies match closely with the prevalent frequencies in the drag coefficient spectrum, and correspond to instabilities due to vortex shedding and jet-flapping. The second unstable CLV captures the subharmonic instability of the shedding frequency. Global linear stability analysis (GLSA) of the time-averaged flow identifies a neutral eigenmode that resembles the leading SPOD mode of the first CLV, with a very similar structure and frequency. However, while GLSA predicts neutrality, Lyapunov analysis reveals that this direction is unstable, exposing the inherent limitations of the GLSA when applied to chaotic flows.
- Open Access
- 7 July 2025
- Published by American Physical Society (APS) in Physical Review Fluids
- Vol. 10 (7)
Spatial structure of the Lamb vector in a separated three-dimensional shear flow around a wing
The Lamb vector appears in the rotational form of the Navier-Stokes equations and is the source of vortical flow nonlinearity that results in the cascade of energy from large to small scales. We investigate the three-dimensional spatial structure of this vector in the near wake of a NACA0018 wing with square wingtip profile at 10 ∘ angle of attack and chord-based Reynolds number of 5000. At these conditions, a large recirculation zone forms in the wake and the flow transitions in the separating shear layer. We analyze the spatial footprint of the three components of the Lamb vector and its divergence (the latter for both the mean and the fluctuating velocity fields). By integrating the vertical component of the Lamb vector across the cross-stream direction (which is equal to the vorticity flux for two-dimensional flows) we obtain the local aerodynamic loading along the wing. The vorticity flux is equal to zero at the trailing edge, thereby confirming the Taylor-Sears condition away from the wing tip. We also apply Helmholtz decomposition to extract the potential and solenoidal parts of the Lamb vector. The former part is dominant in the attached and separated shear layers, while the latter is present downstream of the recirculation zone where the flow is turbulent. We decompose the Euclidean norms of the Lamb vector and its potential part into terms that are determined by the time-average field only, the fluctuating field only, as well as mixed terms. We find that the mixed terms play a dominant role and are the slowest to decay.
- 7 July 2025
- Published by American Physical Society (APS) in Physical Review Fluids
- Vol. 10 (7)
Open AccessSpatial structure of the Lamb vector in a separated three-dimensional shear flow around a wing
The Lamb vector appears in the rotational form of the Navier-Stokes equations and is the source of vortical flow nonlinearity that results in the cascade of energy from large to small scales. We investigate the three-dimensional spatial structure of this vector in the near wake of a NACA0018 wing with square wingtip profile at 10 ∘ angle of attack and chord-based Reynolds number of 5000. At these conditions, a large recirculation zone forms in the wake and the flow transitions in the separating shear layer. We analyze the spatial footprint of the three components of the Lamb vector and its divergence (the latter for both the mean and the fluctuating velocity fields). By integrating the vertical component of the Lamb vector across the cross-stream direction (which is equal to the vorticity flux for two-dimensional flows) we obtain the local aerodynamic loading along the wing. The vorticity flux is equal to zero at the trailing edge, thereby confirming the Taylor-Sears condition away from the wing tip. We also apply Helmholtz decomposition to extract the potential and solenoidal parts of the Lamb vector. The former part is dominant in the attached and separated shear layers, while the latter is present downstream of the recirculation zone where the flow is turbulent. We decompose the Euclidean norms of the Lamb vector and its potential part into terms that are determined by the time-average field only, the fluctuating field only, as well as mixed terms. We find that the mixed terms play a dominant role and are the slowest to decay.
- Open Access
Forecasting the evolution of three-dimensional turbulent recirculating flows from sparse sensor data
A data-driven algorithm is proposed that employs sparse data from velocity and/or scalar sensors to forecast the future evolution of three dimensional turbulent flows. The algorithm combines time-delayed embedding together with Koopman theory and linear optimal estimation theory. It consists of 3 steps; dimensionality reduction (currently POD), construction of a linear dynamical system for current and future POD coefficients and system closure using sparse sensor measurements. In essence, the algorithm establishes a mapping from current sparse data to the future state of the dominant structures of the flow over a specified time window. The method is scalable (i.e.\ applicable to very large systems), physically interpretable, and provides sequential forecasting on a sliding time window of prespecified length. It is applied to the turbulent recirculating flow over a surface-mounted cube (with more than $10^8$ degrees of freedom) and is able to forecast accurately the future evolution of the most dominant structures over a time window at least two orders of magnitude larger that the (estimated) Lyapunov time scale of the flow. Most importantly, increasing the size of the forecasting window only slightly reduces the accuracy of the estimated future states. Extensions of the method to include convolutional neural networks for more efficient dimensionality reduction and moving sensors are also discussed.Open AccessForecasting the evolution of three-dimensional turbulent recirculating flows from sparse sensor data
A data-driven algorithm is proposed that employs sparse data from velocity and/or scalar sensors to forecast the future evolution of three dimensional turbulent flows. The algorithm combines time-delayed embedding together with Koopman theory and linear optimal estimation theory. It consists of 3 steps; dimensionality reduction (currently POD), construction of a linear dynamical system for current and future POD coefficients and system closure using sparse sensor measurements. In essence, the algorithm establishes a mapping from current sparse data to the future state of the dominant structures of the flow over a specified time window. The method is scalable (i.e.\ applicable to very large systems), physically interpretable, and provides sequential forecasting on a sliding time window of prespecified length. It is applied to the turbulent recirculating flow over a surface-mounted cube (with more than $10^8$ degrees of freedom) and is able to forecast accurately the future evolution of the most dominant structures over a time window at least two orders of magnitude larger that the (estimated) Lyapunov time scale of the flow. Most importantly, increasing the size of the forecasting window only slightly reduces the accuracy of the estimated future states. Extensions of the method to include convolutional neural networks for more efficient dimensionality reduction and moving sensors are also discussed. - Open Access
- 9 May 2025
- Published by Cambridge University Press (CUP) in Journal of Fluid Mechanics
- Vol. 1010, A41
Non-normality and small-scale statistics in a three-dimensional, separated shear flow
We analyse the small-scale characteristics, such as enstrophy, total strain and normality/non-normality, in the three-dimensional, separated flow around a NACA 0018 wing using direct numerical simulations. The angle of attack is $10^\circ$ and the Reynolds number (based on the chord length) is $Re_c=5000$ . The role of non-normality is investigated by performing Schur decomposition of the velocity gradient tensor. We also apply the Schur decomposition to derive new expressions for the production of enstrophy and total strain arising from the mean flow inhomogeneity. We focus on two sections of the flow, across the recirculating zone and along the transitioning shear layer, and compare our results with homogeneous isotropic turbulence (HIT). Within the recirculating region, the non-normality index is approximately 0 (and close to the HIT value), indicating almost equal normal and non-normal contributions. However, in the separating layer non-normal effects strongly dominate, especially in the region of kinetic energy growth. Only in the decay region do the values of the non-normality index gradually approximate HIT values. The production of enstrophy due to vortex stretching is dominated by the mixed (interaction) term, where normal strain stretches non-normal vorticity. The same component also dominates the strain self-amplification term. The contributions of different QR regions to the production terms are also examined. Production due to mean strain rate is triggered upstream compared with production due to fluctuating strain fields.- 9 May 2025
- Published by Cambridge University Press (CUP) in Journal of Fluid Mechanics
- Vol. 1010, A41
Open AccessNon-normality and small-scale statistics in a three-dimensional, separated shear flow
We analyse the small-scale characteristics, such as enstrophy, total strain and normality/non-normality, in the three-dimensional, separated flow around a NACA 0018 wing using direct numerical simulations. The angle of attack is $10^\circ$ and the Reynolds number (based on the chord length) is $Re_c=5000$ . The role of non-normality is investigated by performing Schur decomposition of the velocity gradient tensor. We also apply the Schur decomposition to derive new expressions for the production of enstrophy and total strain arising from the mean flow inhomogeneity. We focus on two sections of the flow, across the recirculating zone and along the transitioning shear layer, and compare our results with homogeneous isotropic turbulence (HIT). Within the recirculating region, the non-normality index is approximately 0 (and close to the HIT value), indicating almost equal normal and non-normal contributions. However, in the separating layer non-normal effects strongly dominate, especially in the region of kinetic energy growth. Only in the decay region do the values of the non-normality index gradually approximate HIT values. The production of enstrophy due to vortex stretching is dominated by the mixed (interaction) term, where normal strain stretches non-normal vorticity. The same component also dominates the strain self-amplification term. The contributions of different QR regions to the production terms are also examined. Production due to mean strain rate is triggered upstream compared with production due to fluctuating strain fields. - Open Access
- 25 April 2025
- Published by Elsevier in Journal of Computational Physics
- Vol. 534
An augmented shadowing algorithm for calculating the sensitivity of time-average quantities of chaotic systems
- 25 April 2025
- Published by Elsevier in Journal of Computational Physics
- Vol. 534
Open AccessAn augmented shadowing algorithm for calculating the sensitivity of time-average quantities of chaotic systems
- Open Access
- 23 April 2024
- Published by Cambridge University Press (CUP) in Journal of Fluid Mechanics
- Vol. 985, A27
Resolvent analysis of turbulent flow laden with low-inertia particles
We extend the resolvent framework to two-phase flows with low-inertia particles. The particle velocities are modelled using the equilibrium Eulerian model. We analyse the turbulent flow in a vertical pipe with Reynolds number of $5300$ (based on diameter and bulk velocity), for Stokes numbers $St^+=0-1$ , Froude numbers $Fr_z=-4,-0.4,0.4,4$ and $1/Fr_z = 0$ (gravity omitted). The governing equations are written in input–output form and a singular value decomposition is performed on the resolvent operator. As for single-phase flows, the operator is low rank around the critical layer, and the true response can be approximated using one singular vector. Even with a crude forcing model, the formulation can predict physical phenomena observed in Lagrangian simulations, such as particle clustering and gravitational effects. Increasing the Stokes number shifts the predicted concentration spectra to lower wavelengths; this shift also appears in the direct numerical simulation spectra and is due to particle clustering. When gravity is present, there are two critical layers, one for the concentration field, and one for the velocity field. For upward flow, the peak of concentration fluctuations shifts closer to the wall, in agreement with the literature. We explain this with the aid of the different locations of the two critical layers. Finally, the model correctly predicts the interaction of near-wall vortices with particle clusters. Overall, the resolvent operator provides a useful framework to explain and interpret many features observed in Lagrangian simulations. The application of the resolvent framework to higher $St^+$ flows in combination with Lagrangian simulations is also discussed.- 23 April 2024
- Published by Cambridge University Press (CUP) in Journal of Fluid Mechanics
- Vol. 985, A27
Open AccessResolvent analysis of turbulent flow laden with low-inertia particles
We extend the resolvent framework to two-phase flows with low-inertia particles. The particle velocities are modelled using the equilibrium Eulerian model. We analyse the turbulent flow in a vertical pipe with Reynolds number of $5300$ (based on diameter and bulk velocity), for Stokes numbers $St^+=0-1$ , Froude numbers $Fr_z=-4,-0.4,0.4,4$ and $1/Fr_z = 0$ (gravity omitted). The governing equations are written in input–output form and a singular value decomposition is performed on the resolvent operator. As for single-phase flows, the operator is low rank around the critical layer, and the true response can be approximated using one singular vector. Even with a crude forcing model, the formulation can predict physical phenomena observed in Lagrangian simulations, such as particle clustering and gravitational effects. Increasing the Stokes number shifts the predicted concentration spectra to lower wavelengths; this shift also appears in the direct numerical simulation spectra and is due to particle clustering. When gravity is present, there are two critical layers, one for the concentration field, and one for the velocity field. For upward flow, the peak of concentration fluctuations shifts closer to the wall, in agreement with the literature. We explain this with the aid of the different locations of the two critical layers. Finally, the model correctly predicts the interaction of near-wall vortices with particle clusters. Overall, the resolvent operator provides a useful framework to explain and interpret many features observed in Lagrangian simulations. The application of the resolvent framework to higher $St^+$ flows in combination with Lagrangian simulations is also discussed. - Open Access
- 12 April 2024
- Published by American Physical Society (APS) in Physical Review E
- Vol. 109 (4) , 044208
Uncertainty quantification of time-average quantities of chaotic systems using sensitivity-enhanced polynomial chaos expansion
We consider the effect of multiple stochastic parameters on the time-average quantities of chaotic systems. We employ the recently proposed sensitivity-enhanced generalized polynomial chaos expansion, se-gPC, to quantify efficiently this effect. se-gPC is an extension of gPC expansion, enriched with the sensitivity of the time-averaged quantities with respect to the stochastic variables. To compute these sensitivities, the adjoint of the shadowing operator is derived in the frequency domain. Coupling the adjoint operator with gPC provides an efficient uncertainty quantification algorithm, which, in its simplest form, has computational cost that is independent of the number of random variables. The method is applied to the Kuramoto-Sivashinsky equation and is found to produce results that match very well with Monte Carlo simulations. The efficiency of the proposed method significantly outperforms sparse-grid approaches, such as Smolyak quadrature. These properties make the method suitable for application to other dynamical systems with many stochastic parameters. Published by the American Physical Society 2024
- 12 April 2024
- Published by American Physical Society (APS) in Physical Review E
- Vol. 109 (4) , 044208
Open AccessUncertainty quantification of time-average quantities of chaotic systems using sensitivity-enhanced polynomial chaos expansion
We consider the effect of multiple stochastic parameters on the time-average quantities of chaotic systems. We employ the recently proposed sensitivity-enhanced generalized polynomial chaos expansion, se-gPC, to quantify efficiently this effect. se-gPC is an extension of gPC expansion, enriched with the sensitivity of the time-averaged quantities with respect to the stochastic variables. To compute these sensitivities, the adjoint of the shadowing operator is derived in the frequency domain. Coupling the adjoint operator with gPC provides an efficient uncertainty quantification algorithm, which, in its simplest form, has computational cost that is independent of the number of random variables. The method is applied to the Kuramoto-Sivashinsky equation and is found to produce results that match very well with Monte Carlo simulations. The efficiency of the proposed method significantly outperforms sparse-grid approaches, such as Smolyak quadrature. These properties make the method suitable for application to other dynamical systems with many stochastic parameters. Published by the American Physical Society 2024