In this paper, we examine two widely-used approaches, the polynomial chaos expansion (PCE) and Gaussian process (GP) regression, for the development of surrogate models. The theoretical differences between the PCE and GP approximations are discussed. A state-of-the-art PCE approach is constructed based on high precision quadrature points; however, the need for truncation may result in potential precision loss; the GP approach performs well on small datasets and allows a fine and precise trade-off between fitting the data and smoothing, but its overall performance depends largely on the training dataset. The reproducing kernel Hilbert space (RKHS) and Mercer’s theorem are introduced to form a linkage between the two methods. The theorem has proven that the two surrogates can be embedded in two isomorphic RKHS, by which we propose a novel method named Gaussian process on polynomial chaos basis (GPCB) that incorporates the PCE and GP. A theoretical comparison is made between the PCE and GPCB with the help of the Kullback–Leibler divergence. We present that the GPCB is as stable and accurate as the PCE method. Furthermore, the GPCB is a one-step Bayesian method that chooses the best subset of RKHS in which the true function should lie, while the PCE method requires an adaptive procedure. Simulations of 1D and 2D benchmark functions show that GPCB outperforms both the PCE and classical GP methods. In order to solve high dimensional problems, a random sample scheme with a constructive design (i.e., tensor product of quadrature points) is proposed to generate a valid training dataset for the GPCB method. This approach utilizes the nature of the high numerical accuracy underlying the quadrature points while ensuring the computational feasibility. Finally, the experimental results show that our sample strategy has a higher accuracy than classical experimental designs; meanwhile, it is suitable for solving high dimensional problems.
Detection in high fidelity of tipping points, the emergence of which is often induced by invisible changes in internal structures or/and external interferences, is paramountly beneficial to understanding and predicting complex dynamical systems (CDSs). Detection approaches, which have been fruitfully developed from several perspectives (e.g., statistics, dynamics, and machine learning), have their own advantages but still encounter difficulties in the face of high-dimensional, fluctuating datasets. Here, using the reservoir computing (RC), a recently notable, resource-conserving machine learning method for reconstructing and predicting CDSs, we articulate a model-free framework to accomplish the detection only using the time series observationally recorded from the underlying unknown CDSs. Specifically, we encode the information of the CDS in consecutive time durations of finite length into the weights of the readout layer in an RC, and then we use the learned weights as the dynamical features and establish a mapping from these features to the system’s changes. Our designed framework can not only efficiently detect the changing positions of the system but also accurately predict the intensity change as the intensity information is available in the training data. We demonstrate the efficacy of our supervised framework using the dataset produced by representative physical, biological, and real-world systems, showing that our framework outperforms those traditional methods on the short-term data produced by the time-varying or/and noise-perturbed systems. We believe that our framework, on one hand, complements the major functions of the notable RC intelligent machine and, on the other hand, becomes one of the indispensable methods for deciphering complex systems.
The increasing use of mathematical techniques in scientific research leads to the interdisciplinarity of applied mathematics. This viewpoint is validated quantitatively here by statistical and network analysis on the corpus PNAS 1999–2013. A network describing the interdisciplinary relationships between disciplines in a panoramic view is built based on the corpus. Specific network indicators show the hub role of applied mathematics in interdisciplinary research. The statistical analysis on the corpus content finds that algorithms, a primary topic of applied mathematics, positively correlates, increasingly co-occurs, and has an equilibrium relationship in the long-run with certain typical research paradigms and methodologies. The finding can be understood as an intrinsic cause of the interdisciplinarity of applied mathematics.
The TH-2 satellite system, including the TH-2A and TH-2B, is the first distributed interferometric synthetic aperture radar (InSAR) satellite system in China. During the in-orbit operation, the TH-2A satellite should perform three maneuvers per day to keep the formation flying geometry. We estimate those maneuvers in the precise orbit determination (POD) by the GPS and BDS2 measurements on board, respectively. The residuals of the POD show that the effects caused by orbital maneuvers can be well eliminated for both the GPS and BDS2 data. The precision of the BDS2-based POD is better than 8.0 cm in the three-dimensional direction (3D) compared with the orbit derived from the GPS observations. Such a precision level of the satellite orbit satisfies the InSAR mission requirement of the TH-2. In addition, the relative error of velocity changes is employed to evaluate the maneuver estimations by the POD using the regional navigation system of BDS2. The results show that the relative error of velocity changes between the GPS- and BDS2-based POD is less than 7.0%, which indicates that the maneuver performance extracted from the regional BDS2 data is as good as that extracted from the global GPS data. In the GNSS fused processing, we found that the independent receiver clock offsets should be taken into account, since the time tag corrections for the GPS and BDS2 observations collected on the TH-2 spaceborne receivers were different. The precision of the GPS and BDS2 (GC) combined single point positioning (SPP) can be improved by 12–14% compared with the GPS-only solution when the position dilution of precision (PDOP) of GPS exceeds three. The overlap comparisons of the GC combined orbits show that the internal orbit precision of the TH-2 satellites is better than 0.7 cm. However, the improvement of the GC combined POD result is only 3–4% with respect to the GPS-only solution, which is limited to the precision of the precise orbit and clock products of BDS2 at the present stage.