Processing math: 100%
Research article

SVD-Krylov based techniques for structure-preserving reduced order modelling of second-order systems

  • Correction on: Mathematical Modelling and Control 1: 134-135
  • Received: 21 March 2021 Accepted: 16 June 2021 Published: 22 June 2021
  • We introduce an efficient structure-preserving model-order reduction technique for the large-scale second-order linear dynamical systems by imposing two-sided projection matrices. The projectors are formed based on the features of the singular value decomposition (SVD) and Krylov-based model-order reduction methods. The left projector is constructed by utilizing the concept of the observability Gramian of the systems and the right one is made by following the notion of the interpolation-based technique iterative rational Krylov algorithm (IRKA). It is well-known that the proficient model-order reduction technique IRKA cannot ensure system stability, and the Gramian based methods are computationally expensive. Another issue is preserving the second-order structure in the reduced-order model. The structure-preserving model-order reduction provides a more exact approximation to the original model with maintaining some significant physical properties. In terms of these perspectives, the proposed method can perform better by preserving the second-order structure and stability of the system with minimized H2-norm. Several model examples are presented that illustrated the capability and accuracy of the introducing technique.

    Citation: Md. Motlubar Rahman, Mahtab Uddin, M. Monir Uddin, L. S. Andallah. SVD-Krylov based techniques for structure-preserving reduced order modelling of second-order systems[J]. Mathematical Modelling and Control, 2021, 1(2): 79-89. doi: 10.3934/mmc.2021006

    Related Papers:

    [1] Md. Motlubar Rahman, Mahtab Uddin, M. Monir Uddin, L. S. Andallah . Correction: SVD-Krylov based techniques for structure-preserving reduced order modelling of second-order systems. Mathematical Modelling and Control, 2021, 1(3): 134-135. doi: 10.3934/mmc.2021011
    [2] Naveen Kumar, Km Shelly Chaudhary . Position tracking control of nonholonomic mobile robots via $ H_\infty $-based adaptive fractional-order sliding mode controller. Mathematical Modelling and Control, 2025, 5(1): 121-130. doi: 10.3934/mmc.2025009
    [3] Lei Wang, Xinyun Liu, Ting Li, Jiandong Zhu . Skew-symmetric games and symmetric-based decomposition of finite games. Mathematical Modelling and Control, 2022, 2(4): 257-267. doi: 10.3934/mmc.2022024
    [4] Xiaomeng Wei, Haitao Li, Guodong Zhao . Kronecker product decomposition of Boolean matrix with application to topological structure analysis of Boolean networks. Mathematical Modelling and Control, 2023, 3(4): 306-315. doi: 10.3934/mmc.2023025
    [5] Xueling Fan, Ying Li, Wenxv Ding, Jianli Zhao . $ \mathcal{H} $-representation method for solving reduced biquaternion matrix equation. Mathematical Modelling and Control, 2022, 2(2): 65-74. doi: 10.3934/mmc.2022008
    [6] Alexandru Hofman, Radu Precup . On some control problems for Kolmogorov type systems. Mathematical Modelling and Control, 2022, 2(3): 90-99. doi: 10.3934/mmc.2022011
    [7] Rashid Jan, Normy Norfiza Abdul Razak, Sania Qureshi, Imtiaz Ahmad, Salma Bahramand . Modeling Rift Valley fever transmission: insights from fractal-fractional dynamics with the Caputo derivative. Mathematical Modelling and Control, 2024, 4(2): 163-177. doi: 10.3934/mmc.2024015
    [8] Anil Chavada, Nimisha Pathak . Transmission dynamics of breast cancer through Caputo Fabrizio fractional derivative operator with real data. Mathematical Modelling and Control, 2024, 4(1): 119-132. doi: 10.3934/mmc.2024011
    [9] Anil Chavada, Nimisha Pathak, Sagar R. Khirsariya . A fractional mathematical model for assessing cancer risk due to smoking habits. Mathematical Modelling and Control, 2024, 4(3): 246-259. doi: 10.3934/mmc.2024020
    [10] Jianhua Sun, Ying Li, Mingcui Zhang, Zhihong Liu, Anli Wei . A new method based on semi-tensor product of matrices for solving reduced biquaternion matrix equation $ \sum\limits_{p = 1}^l A_pXB_p = C $ and its application in color image restoration. Mathematical Modelling and Control, 2023, 3(3): 218-232. doi: 10.3934/mmc.2023019
  • We introduce an efficient structure-preserving model-order reduction technique for the large-scale second-order linear dynamical systems by imposing two-sided projection matrices. The projectors are formed based on the features of the singular value decomposition (SVD) and Krylov-based model-order reduction methods. The left projector is constructed by utilizing the concept of the observability Gramian of the systems and the right one is made by following the notion of the interpolation-based technique iterative rational Krylov algorithm (IRKA). It is well-known that the proficient model-order reduction technique IRKA cannot ensure system stability, and the Gramian based methods are computationally expensive. Another issue is preserving the second-order structure in the reduced-order model. The structure-preserving model-order reduction provides a more exact approximation to the original model with maintaining some significant physical properties. In terms of these perspectives, the proposed method can perform better by preserving the second-order structure and stability of the system with minimized H2-norm. Several model examples are presented that illustrated the capability and accuracy of the introducing technique.



    Mathematical modeling of physical systems plays a significant role in assessing, simulate, control, or optimize the implementation of real-world problems in technological development. It is an alternative approach to avoid extensive laboratory works engaged with a great deal of equipment. On the other hand, mathematical modeling can save not only time or labor but also help to enhance the performance of the physical system. The models are originated in different ways, such as finite element method (FEM) or finite difference method (FDM) discretization, and often represented by linear time-invariant (LTI) continuous-time system. The second-order LTI continuous-time system can be represented in state-state form as

    M¨ξ(t)+D˙ξ(t)+Kξ(t)=Hu(t),y(t)=Lξ(t)+Dau(t), (1.1)

    where M,D,KRn×n are time-invariant matrices, HRn×p is the input matrix describing the external access to the system, LRm×n represents the output of the measurement and DaRm×p is the direct feed-through of the system. The number of states or dimensions of the system is n while ξ(t)Rn is the vector of states, u(t)Rp is the vector of control input and y(t)Rm is the measured output of the system. The input and output of the system are defined in continuous-time over the interval [0,) and thus the system is known as a continuous-time system of second-order. If M=I, then the system is called a standard state-space system, or if M is an invertible matrix then the system can also be converted into a standard state-space system. If p=m=1, the system is called single-input single-output (SISO) system, otherwise, it is called the multi-input multi-output (MIMO) system. In the MIMO system, we consider p,m<<n, i.e., the number of input and output of the system is much less than the number of states. Applying Laplace transformation, the transfer function of second-order LTI continuous system can be found as

    G(s)=L(s2M+sD+K)1H+Da;sC. (1.2)

    Many physical phenomena of science and engineering can be modeled as second-order systems (see e.g., [1,2,3]). For example, in mechanics, mechatronics [4], and multi-body dynamics [5], where the velocity is taken into account in the modeling, and thus the acceleration becomes part of the system. In mechanics, usually, the matrices M, D, and K describe the mass, damping, and stiffness distributions, respectively, and the vector ξ(t) is known as mechanical displacement [6]. Such systems also appear in electrical engineering when RLC circuits are designed for nodal analysis [7]. There the matrices M, D, and K are called the conductance, capacitance, and susceptance matrices, respectively, and the vector ξ(t) is denoted as electric charge.

    To generate an exact model of a physical system, through discretization, a sufficient number of grid points needs to create because the system may obstruct by several bodies with factious devices. As a result, the size of the developing model usually becomes too large and sparse, and additionally well-structured also. A large-scale system demands extraordinary computational effort to deal with and consume additional computer memory to store it. It also prohibits rapid simulation processes that are obligatory in many installations. To avoid these types of complications, reducing the size of the system becomes unavoidable in many cases. The technique to reduce the size or dimension of a model is known as model order reduction (MOR) [8,9,10]. The r-dimensional reduced-order model (ROM) of (1.1) can be written as

    ˆM¨ˆξ(t)+ˆD˙ˆξ(t)+ˆKˆξ(t)=ˆHu(t),ˆy(t)=ˆLˆξ(t)+ˆDau(t). (1.3)

    where ˆM,ˆD,ˆKRr×r, ˆHRr×p, ˆLRm×r, ˆDa=Da and rn. The target is to minimize the value of r by the trial and error basis such that system (1.3) approximates to the system (1.1) by conserving the system properties invariant. The transfer function corresponding to the ROM (1.3) is denoted by ˆG(s) and is defined as

    ˆG(s)=ˆL(s2ˆM+sˆD+ˆK)1ˆH+ˆDa;sC. (1.4)

    The aspect of model order reduction is to reduce the unwanted enormous computational interruptions in analyzing or improving the performance in the implementation of modern technology. Among several model reduction techniques, SVD based balanced truncation (BT) [11,12] and Krylov-based techniques iterative rational Krylov algorithm (IRKA) [13,14] are well-known and efficient in this arena. Both techniques have some advantages or disadvantages such as IRKA is computationally cheap, whereas BT is comparatively expensive, BT has a priori error bound, in contrast, IRKA has no error bound. Also, BT preserves the stability of the system, contrariwise IRKA does not depend on it. By going beyond these issues, combining the features of both these techniques S. Gugercin in [15] developed an efficient way for model-order reduction of large-scale linear dynamical systems that assures the system stability asymptotically with minimized H2 approximation. We propose a two-sided projection-based model-order reduction method iterative SVD-Krylov algorithm (ISKA) for the second-order linear dynamical system. The method is susceptible to provide the best approximation to the full model by preserving the second-order structure and stability of the system with minimized H2 norm.

    Consider the k-dimensional linear dynamical system represents in state-space form of first-order ordinary differential equation as

    E˙x(t)=Ax(t)+Bu(t),x(t0)=x0,tt0,y(t)=Cx(t)+Dau(t), (2.1)

    where ERk×k is non-singular, ARk×k, BRk×p, CRm×k and DaRm×p.

    However, the second-order system (1.1) can be converted into an equivalent first-order form (2.1) in several ways [16]. Due to the structural requirement of the system, a feasible first-order representation of the second-order system is needed to be constructed. To design a suitable arrangement of the second-order system (1.1), we are choosing the following first-order form

    x(t)=[ξ(t)T ˙ξ(t)T]T, E=[I00M], A=[0IKD],B=[0H], C=[L0], and Da=Da. (2.2)

    The systematize observability Lyapunov equation of (2.1) is

    ATQE+ETQA=CTC. (2.3)

    Since the direct computation of the observability Gramian by solving equation (2.3) is almost impossible for large-scale settings, one needs to find the corresponding Gramian factor by any suitable techniques, for instance, Low-rank Cholesky-factor based Alternating Direction Implicit (LRCF-ADI) method as in [17,18,19]. If Zq the observability Gramian factor, then the Gramian, Q=ZqZTq can be treated as the approximate solution of the Lyapunov equation (2.3). The techniques for finding Zq of first-order system is provided in Algorithm 1.

    Algorithm 1: First-order LRCF-ADI [17,19]
    Input: E,A,B,C, τ (tolerance), imax (number of iterations) and shift parameters {μj}imaxj=1.
    Output: Low-rank Cholesky-factor Zq such that QZqZTq.
    1 Consider at i=1, Z0=[] and W0=CT.
    2 while Wi1WTi1τ or iimax do

    14 end while

     | Show Table
    DownLoad: CSV

    Now consider the computationally feasible r-dimensional reduced order model of (2.1) is given by

    ˆE˙ˆx(t)=ˆAˆx(t)+ˆBˆu(t),ˆy(t)=ˆCˆx(t)+ˆDaˆu(t), (2.4)

    where ˆERr×r, ˆARr×k, ˆBRr×p, ˆCRm×r and ˆDaRm×p.

    The reduced coefficient matrices of (2.4) is formed by the following way

    ˆE=WTEV,ˆA=WTAV,ˆB=WTB,ˆC=CV,ˆDa=Da. (2.5)

    Here the projector V is constructed by following the prominent Krylov-based interpolatory MOR techniques IRKA in [14,20] as

    V=[(α1EA)1Bb1,,(αrEA)1Bbr], (2.6)

    in which {αi}ri=1 and {bi}ri=1 are the assuming interpolation points and tangential directions respectively, where r is the size the desired ROM. The other projector W is formed with observability Gramian Q according to the SVD-based MOR techniques in [15,21,22] as

    W=QV(VTQV)1. (2.7)

    The detailed procedure of computing the ROMs (2.4) of first-order system is illustrated in Algorithm 2. In [15,Theorem 3.2] S. Gugercin showed that the ROMs (2.4) achieved by Algorithm 2 maintaining the stability of the system asymptotically as the original system.

    Algorithm 2: First-order ISKA [15]
    Input: E,A,B,C,Da,andZq (from Algorithm 1).
    Output: ˆE,ˆA,ˆB,ˆC, ^Da:=Da.
    1 Consider the initial assumptions for the the interpolation points {αi}ri=1 and the tangential directions {bi}ri=1.
    2 Construct V=[(α1EA)1Bb1,,(αrEA)1Bbr]
    3 Compute Q=ZqZTq and construct W=QV(VTQV)1.
    4 while (not converged) do

    11 end while
    12 Find the reduced-order matrices by repeating Step-5.

     | Show Table
    DownLoad: CSV

    The main intention of this work is to reduce the dimension of the second-order system (1.1) by keeping the system structure invariant through the ISKA approach. To do this, it is essential to modify some steps of first-order ISKA and LRCF-ADI algorithms in terms of second-order matrices of (1.1).

    The LRCF-ADI method of first-order was discussed in [19,23,24]. The modification of the LRCF-ADI algorithm for the structure-preserving second-order form can be derived as follows.

    Let us consider Vi=[V(1)iV(2)i] and Wi=[W(1)iW(2)i], then for the first iteration of the Step-3 of Algorithm 1 can be written as

    (AT+μ1ET)V1=CT,or,([0IKD]T+μ1[I00M]T)[V(1)1V(2)1]=[L0]T. (3.1)

    From (3.1) we get

    V(1)1=(μ1MTDT)V(2)1,V(2)1=(KT+μ1DTμ21MT)1LT. (3.2)

    As a consequence, for i2, the next ith iterations take the forms

    V(1)i=(μiMTDT)V(2)i,V(2)i=(KT+μiDTμ2iMT)1W(1)i1. (3.3)

    If the shift parameter has no imaginary part, then the Step-6 of Algorithm 1 can be written as

    Wi=Wi12μiETVi,or,[W(1)iW(2)i]=[W(1)i1W(2)i1]2μi[I00M]T[V(1)iV(2)i], (3.4)

    which implies

    W(1)i=W(1)i12μiV(1)i,W(2)i=W(2)i12μiMTV(2)i. (3.5)

    and if it contains an imaginary part, then for δi=Re(μi)Im(μi), the Step-10 of Algorithm 1 can be expressed as

    Wi+1=Wi14Re(μi)ET[Re(Vi)+δiIm(Vi)],or,[W(1)i+1W(2)i+1]=[W(1)i1W(2)i1]4Re(μi)[I00M]T(Re[V(1)iV(2)i]+δiIm[V(1)iV(2)i]). (3.6)

    This results in

    W(1)i+1=W(1)i14Re(μi)[Re(V(1)i)+δiIm(V(1)i)],W(2)i+1=W(2)i14Re(μi)MT[Re(V(2)i)+δiIm(V(2)i)]. (3.7)

    The re-organized second-order structure-preserving form of LRCF-ADI is exhibited in Algorithm 3.

    Algorithm 3: Second-order structure-preserving LRCF-ADI
    Input: M,D,K,H,L, τ (tolerance), imax (number of iterations) and shift parameters {μj}imaxj=1.
    Output: Low-rank Cholesky-factor Zq such that QZqZTq.
    1 Consider at i=1, Z0=[] and W0=CT from (2.2).
    2 while Wi1WTi1τ or iimax do

    14 end while

     | Show Table
    DownLoad: CSV

    Algorithm 2 needs to reform in the structure-preserving shape with the system matrices of (1.1). In the Step-2 of this algorithm, projector V needs to be re-structured utilizing the second-order matrices. Let us consider the ith iteration of V be expressed as Vi and due to the structure of the system, it can be partitioned as Vi=[V(1)iV(2)i], then for the MIMO dynamical systems, it can be configured as

    (αiEA)Vi=Bbi,or,[αiIIKαiM+D][V(1)iV(2)i]=[0H]bi. (3.8)

    Equation (3.8) implies to

    V(1)i=(α2iM+αiD+K)1Hbi,V(2)i=αiV(1)i. (3.9)

    Likewise for the SISO case, without considering the tangential direction bi, (3.9) can be written as

    V(1)i=(α2iM+αiD+K)1H,V(2)i=αiV(1)i. (3.10)

    Now, the error system associated with the reduced-order model (1.3) of the considering second-order system (1.1) by maintaining the first-order representation (2.2) has the form

    Gerr=G(s)ˆG(s)=Cerr(sEerrAerr)1Berr, (3.11)

    where G(s) and ˆG(s) are defined in (1.2) and (1.4), respectively. In (3.11), we have constituted

    Eerr=[E00ˆE], Aerr=[A00ˆA],Berr=[BˆB], and Cerr=[CˆC]. (3.12)

    Here, E,A,B and C are the matrices provided in the first-order representation (2.2) of the second-order system (1.1) and also ˆE,ˆA,ˆB and ˆC have the following compositions

    ˆE=[I00ˆM], ˆA=[0IˆKˆD],ˆB=[0ˆH], ˆC=[ˆL0]. (3.13)

    The observability Lyapunov equation corresponding to the Graminan Qerr of the error system (3.11) is

    ATerrQerrEerr+ETerrQerrAerr+CTerrCerr=0. (3.14)

    Authors in [25] evolved an efficient approach to estimate the H2 norm of the error system (3.11) as

    GerrH2=trace(BTerrQerrBerr)=G(s)2H2+ˆG(s)2H2+2trace(BTQsˆB). (3.15)

    Here, G(s)H2 is the H2 norm of the full model which we need to evaluate at one time in computation but this is infeasible to investigate for a large-scale system by any direct solver. Suppose Zq is the low-rank Gramian factor of the Gramian Qerr of the Lyapunov equation (3.14) then

    G(s)2H2=trace(BTQerrB)=trace(HT(ZqZTq)H), (3.16)

    and that can be successfully determined by Algorithm 3.

    Again, the H2 norm of the reduced-order model, ˆG(s)H2 can be enumerated by the Gramian ˆQ of the low-rank Lyapunov equation

    ˆATˆQˆE+ˆETˆQˆA=ˆCTˆC (3.17)

    that consists of reduced-order matrices. Due to the small size of these matrices, the Lyapunov equation (3.17) is solvable by the MATLAB library command lyap.

    Algorithm 4: Second-order structure-preserving ISKA with H2 norm
    Input: M,D,K,H,L,Da,andZq (from Algorithm 3).
    Output: ˆM,ˆD,ˆK,ˆH,ˆL, ˆDa:=Da, and GerrH2.
    1 Consider the initial assumptions for the the interpolation points {αi}ri=1 and the tangential directions {bi}ri=1.
    2 Construct V=[V1,V2,,Vr] using (3.9) or (3.10).
    3 Compute Q=ZqZTq and construct W=QV(VTQV)1.
    4 while (not converged) do

    11 end while
    12 Find the reduced matrices by repeating Step-5.
    13 Solve equation (3.17) by MATLAB library command lyap.
    14 Solve equation (3.18) by Algorithm 4 of [26].
    15 Estimate GerrH2 by the equation (3.15).

     | Show Table
    DownLoad: CSV

    Finally, trace(BTQsˆB) can be measured by the low-rank Gramian Qs of the sparse-dense Sylvester equation

    ATQsˆE+ETQsˆA=CTˆC, (3.18)

    that can be efficiently solved by the techniques provided in Algorithm 4 of [26].

    The second-order structure-preserving modified form of Algorithm 2 with the computation technique of H2 norm is summarised in Algorithm 4.

    The author in [3] showed that the second-order systems are equivalent to the corresponding first-order systems, hence system (1.3) is equivalent to (2.4). Thus, according to [15] the reduced-order system (1.3) achieved by Algorithm 4 is asymptotically stable.

    The proposed method ISKA is validated by applying to some data generated from real-world problems. International Space Station Model (ISSM), Clamped Beam Model (CBM), Scalable Oscillator Model (SOM), and Butterfly Gyro Model (BGM) are under our attention. The numerical computations are carried out with MATLAB R2015a (8.5.0.197613) on a board with 4×IntelCoreTMi5-6200U CPU incorporating a 2.30 GHz clock speed and 16 GB RAM.

    Table 1 displays the dimensions of the discussing models, their types with analogous input-output structures, and the size of the corresponding ROMs gain by the developed technique ISKA as illustrated in Algorithm 4. Detailed of those models are available on the web-page for the Oberwolfach Benchmark Collection*.

    Table 1.  Model examples with input-output structures.
    Model Full model (n) Input/Output ROM (r)
    ISSM 270 3/3 20
    CBM 348 1/1 30
    SOM 9001 1/1 50
    BGM 17361 12/12 70

     | Show Table
    DownLoad: CSV

    *https://sparse.tamu.edu/Oberwolfach

    Figure 1a, Figure 2a, Figure 3a and Figure 4a display the well-matching of the transfer functions of the full models with analogous ROMs in the desired dimensions. Figure 1b, Figure 2b, Figure 3b and Figure 4b depict the absolute errors in computing ROMs of the relating second-order models, whereas Figure 1c, Figure 2c, Figure 3c and Figure 4c illustrate the corresponding relative errors in attaining the ROMs.

    Figure 1.  Comparison of full model and ROM of the model ISSM.
    Figure 2.  Comparison of full model and ROM of the model CBM.
    Figure 3.  Comparison of full model and ROM of the model SOM.
    Figure 4.  Comparison of full model and ROM of the model BGM.

    From the displaying figures achieved by ISKA, it is conspicuous that the proposed method is sufficiently efficient and robust for the target models in finding ROMs of the second-order systems by conserving the second-order structure.

    In comparison with the previous work [25], it is noticed that Algorithm 4 can generate a better approximation to the full model. Since the graphical presentations are almost the same, to compendious the article, we are avoiding the graphical comparison here but we will present the H2-norm comparisons next.

    Sub-figures of Figure 5 exhibit the step-responses of the ROMs of the target models originate from Algorithm 4. The horizontal axis depicts the time required for the step-response to be converged to the equilibrium, whereas the vertical axis is for the amplitude. The ROM of BGM converged within a fraction of a second, and the ROMs of ISSM, CBM, and SOM need 100, 500, and 10000 seconds, respectively, to be converged.

    Figure 5.  Step-responses of the ROMs.

    It is perceivable from Figure 5 that all the ROMs computed by Algorithm 4 of the considering models are stable after a certain period of time, i.e, the ROMs are asymptotically stable.

    Table 2 represents the comparisons of the H2 error norm of the ROMs achieved by IRKA and our newly developed technique as represents in Algorithm 4. Here, we compared the H2 error norm estimated by ISKA in the present work with that of the previous work for concerning models in [25] by IRKA.

    Table 2.  Display of H2 error norm of the ROMs.
    Model IRKA ISKA
    Position Velocity
    ISSM 1.2×106 5.1×106 4.6×107
    CBM 1.1×103 3.6×103 2.3×104
    SOM 1.8×101 3.6×101 3.4×102
    BGM 6.6×1012 7.8×1012 1.9×1012

     | Show Table
    DownLoad: CSV

    It has been observed that the developed approach ISKA is well-efficient to minimize the H2 error norms of all the considering models in comparison to IRKA and that ascertain the proposed technique can provide a better approximation to the original model by conserving system properties.

    Moreover, from Table 2, it is evident that the MIMO systems are more feasible in terms of H2 norm optimality in comparison to the SISO systems.

    We have introduced a two-sided projection-based structure-preserving model-order reduction approach for the second-order linear dynamical systems. Here we have following the features of the SVD and Krylov-based model reduction methods. We have successfully modified the LRCF-ADI and ISKA algorithms of first-order to preserve the second-order structure in the reduced-order model. Structure-preservation of the system is significant for getting a better approximation to the full model. It conserves some fundamental physical properties of the system that is essential for more exploration of the system. The SVD-based method has a priori error bound but it has computational complexities, and the Krylov-based technique is computationally efficient but it has neither an error bound nor the stability guarantee. The numerical investigations of the proposed method on some real-world models manifest that it is productive to provide the reduced-order model by preserving the second-order structure and system stability with minimized H2 norm of the reduced system.

    This research is funded by University Grant Commission (UGC), Bangladesh. The reference of the funding is ugc/fellowship/1.157/ph.d/2015/part-2 and has been started from December 02, 2018.

    The authors declare that they have no conflicts of interest to this work.



    [1] V. I. Arnol'd, Mathematical methods of classical mechanics, Springer Science & Business Media, 60 (2013).
    [2] R. Riaza, Differential-algebraic systems: analytical aspects and circuit applications, World Scientific, (2008).
    [3] M. M. Uddin, Computational Methods for Approximation of Large-scale Dynamical Systems, Chapman and Hall/CRC, (2019).
    [4] M. M. Rahman, M. M. Uddin, L. Andallah, M. Uddin, Tangential interpolatory projections for a class of second-order index-1 descriptor systems and application to mechatronics, Production Engineering, 15 (2021), 9–19. doi: 10.1007/s11740-020-00995-4
    [5] E. Eich-Soellner, C. Führer, Numerical methods in multibody dynamics, Springer, 45 (1998).
    [6] R. R. Craig Jr, A. J. Kurdila, Fundamentals of structural dynamics, John Wiley & Sons, (2006).
    [7] B. Yan, S. X.-D. Tan, B. McGaughy, Second-order balanced truncation for passive-order reduction of rlck circuits, IEEE Transactions on Circuits and Systems II: Express Briefs, 55 (2008), 942–946. doi: 10.1109/TCSII.2008.925655
    [8] A. C. Antoulas, Approximation of large-scale dynamical systems, SIAM, (2005).
    [9] P. Benner, V. Mehrmann, D. C. Sorensen, Dimension reduction of large-scale systems, Springer, 45 (2005).
    [10] M. M. Uddin, Computational methods for model reduction of large-scale sparse structured descriptor systems, (2015).
    [11] B. Moore, Principal component analysis in linear systems: Controllability, observability, and model reduction, IEEE transactions on automatic control, 26 (1981), 17–32. doi: 10.1109/TAC.1981.1102568
    [12] M. S. Tombs, I. Postlethwaite, Truncated balanced realization of a stable non-minimal state-space system, International Journal of Control, 46 (1987), 1319–1330. doi: 10.1080/00207178708933971
    [13] E. Grimme, Krylov projection methods for model reduction, Ph.D. dissertation, University of Illinois at Urbana Champaign, (1997).
    [14] S. Gugercin, A. C. Antoulas, C. Beattie, H_2 model reduction for large-scale linear dynamical systems, SIAM journal on matrix analysis and applications, 30 (2008), 609–638. doi: 10.1137/060666123
    [15] S. Gugercin, An iterative svd-krylov based method for model reduction of large-scale dynamical systems, Linear Algebra and its Applications, 428 (2008), 1964–1986. doi: 10.1016/j.laa.2007.10.041
    [16] F. Tisseur, K. Meerbergen, The quadratic eigenvalue problem, SIAM review, 43 (2001), 235–286. doi: 10.1137/S0036144500381988
    [17] P. Benner, J. Saak, Efficient balancing based mor for second order systems arising in control of machine tools, Proceedings of the MathMod, (2009).
    [18] M. S. Hossain, S. G. Omar, A. Tahsin, E. H. Khan, Efficient system reduction modeling of periodic control systems with application to circuit problems, in 2017 4th International Conference on Advances in Electrical Engineering (ICAEE), IEEE, (2017), 259–264.
    [19] J.-R. Li, Model reduction of large linear systems via low rank system gramians, Ph.D. dissertation, Massachusetts Institute of Technology, (2000).
    [20] S. A. Wyatt, Issues in interpolatory model reduction: Inexact solves, second-order systems and daes, Ph.D. dissertation, Virginia Tech, (2012).
    [21] M. Khatibi, H. Zargarzadeh, M. Barzegaran, Power system dynamic model reduction by means of an iterative svd-krylov model reduction method, in 2016 IEEE Power & Energy Society Innovative Smart Grid Technologies Conference (ISGT), IEEE, (2016), 1–6.
    [22] S. Li, J. Trevelyan, Z. Wu, H. Lian, D. Wang, W. Zhang, An adaptive svd–krylov reduced order model for surrogate based structural shape optimization through isogeometric boundary element method, Computer Methods in Applied Mechanics and Engineering, 349 (2019), 312–338. doi: 10.1016/j.cma.2019.02.023
    [23] A. Lu, E. L. Wachspress, Solution of lyapunov equations by alternating direction implicit iteration, Computers & Mathematics with Applications, 21 (1991), 43–58.
    [24] P. Benner, J.-R. Li, T. Penzl, Numerical solution of large-scale lyapunov equations, riccati equations, and linear-quadratic optimal control problems, Numerical Linear Algebra with Applications, 15 (2008), 755–777. doi: 10.1002/nla.622
    [25] M. M. Rahman, M. M. Uddin, L. S. Andallah, M. Uddin, Interpolatory projection techniques for H2 optimal structure-preserving model order reduction of second-order systems, Advances in Science, Technology and Engineering Systems Journal, 5 (2020), 715–723. doi: 10.25046/aj050485
    [26] P. Benner, M. Köhler, J. Saak, Sparse-dense sylvester equations in h2-model order reduction, (2011).
  • This article has been cited by:

    1. Md. Motlubar Rahman, Mahtab Uddin, M. Monir Uddin, L. S. Andallah, Correction: SVD-Krylov based techniques for structure-preserving reduced order modelling of second-order systems, 2021, 1, 2767-8946, 134, 10.3934/mmc.2021011
    2. Mahtab Uddin, M. Monir Uddin, Md. Abdul Hakim Khan, SVD-Krylov-based sparsity-preserving techniques to optimally stabilize the incompressible Navier–Stokes flows, 2023, 2195-268X, 10.1007/s40435-023-01159-9
    3. Mahtab Uddin, M. Monir Uddin, Md. Abdul Hakim Khan, Qing-Wen Wang, SVD‐Krylov–Based Structure‐Preserving Techniques for Approximation of a Class of Second‐Order Index‐1 Descriptor Systems, 2025, 2025, 1110-757X, 10.1155/jama/5976163
  • Reader Comments
  • © 2021 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(3813) PDF downloads(138) Cited by(3)

Figures and Tables

Figures(5)  /  Tables(2)

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog