Loading [MathJax]/jax/element/mml/optable/Latin1Supplement.js
Research article

Zeroing neural network model for solving a generalized linear time-varying matrix equation

  • The time-varying solution of a class generalized linear matrix equation with the transpose of an unknown matrix is discussed. The computation model is constructed and asymptotic convergence proof is given by using the zeroing neural network method. Using an activation function, the predefined-time convergence property and noise suppression strategy are discussed. Numerical examples are offered to illustrate the efficacy of the suggested zeroing neural network models.

    Citation: Huamin Zhang, Hongcai Yin. Zeroing neural network model for solving a generalized linear time-varying matrix equation[J]. AIMS Mathematics, 2022, 7(2): 2266-2280. doi: 10.3934/math.2022129

    Related Papers:

    [1] Guijun Xing, Huatao Chen, Zahra S. Aghayan, Jingfei Jiang, Juan L. G. Guirao . Tracking control for a class of fractional order uncertain systems with time-delay based on composite nonlinear feedback control. AIMS Mathematics, 2024, 9(5): 13058-13076. doi: 10.3934/math.2024637
    [2] Jingjing Yang, Jianqiu Lu . Stabilization in distribution of hybrid stochastic differential delay equations with Lévy noise by discrete-time state feedback controls. AIMS Mathematics, 2025, 10(2): 3457-3483. doi: 10.3934/math.2025160
    [3] Kareem Alanazi, Omar Naifar, Raouf Fakhfakh, Abdellatif Ben Makhlouf . Innovative observer design for nonlinear systems using Caputo fractional derivative with respect to another function. AIMS Mathematics, 2024, 9(12): 35533-35550. doi: 10.3934/math.20241686
    [4] Junsoo Lee, Wassim M. Haddad . On finite-time stability and stabilization of nonlinear hybrid dynamical systems. AIMS Mathematics, 2021, 6(6): 5535-5562. doi: 10.3934/math.2021328
    [5] Bin Hang, Weiwei Deng . Finite-time adaptive prescribed performance DSC for pure feedback nonlinear systems with input quantization and unmodeled dynamics. AIMS Mathematics, 2024, 9(3): 6803-6831. doi: 10.3934/math.2024332
    [6] Miao Xiao, Zhe Lin, Qian Jiang, Dingcheng Yang, Xiongfeng Deng . Neural network-based adaptive finite-time tracking control for multiple inputs uncertain nonlinear systems with positive odd integer powers and unknown multiple faults. AIMS Mathematics, 2025, 10(3): 4819-4841. doi: 10.3934/math.2025221
    [7] Tan Zhang, Pianpian Yan . Asymmetric integral barrier function-based tracking control of constrained robots. AIMS Mathematics, 2024, 9(1): 319-339. doi: 10.3934/math.2024019
    [8] Liping Yin, Yangyu Zhu, Yangbo Xu, Tao Li . Dynamic optimal operational control for complex systems with nonlinear external loop disturbances. AIMS Mathematics, 2022, 7(9): 16673-16691. doi: 10.3934/math.2022914
    [9] Taewan Kim, Jung Hoon Kim . A new optimal control approach to uncertain Euler-Lagrange equations: H disturbance estimator and generalized H2 tracking controller. AIMS Mathematics, 2024, 9(12): 34466-34487. doi: 10.3934/math.20241642
    [10] Zhiqiang Chen, Alexander Yurievich Krasnov . Disturbance observer based fixed time sliding mode control for a class of uncertain second-order nonlinear systems. AIMS Mathematics, 2025, 10(3): 6745-6763. doi: 10.3934/math.2025309
  • The time-varying solution of a class generalized linear matrix equation with the transpose of an unknown matrix is discussed. The computation model is constructed and asymptotic convergence proof is given by using the zeroing neural network method. Using an activation function, the predefined-time convergence property and noise suppression strategy are discussed. Numerical examples are offered to illustrate the efficacy of the suggested zeroing neural network models.



    Unless otherwise stated, X will stand for an infinite Tychonoff space. We assume that all linear spaces are over the field R of real numbers and all locally convex spaces are Hausdorff. We denote by Cp(X) the ring C(X) of real-valued continuous functions on X endowed with the pointwise topology τp, and by Ck(X) the space C(X) equipped with the compact-open topology τk. The subspace of C(X) of uniformly bounded functions is represented by Cb(X). The topological dual of Cp(X) is denoted by L(X), or by Lp(X) when equipped with the weak* topology σ(L(X),C(X)). If we provide L(X) with the strong topology β(L(X),C(X)), we shall write Lβ(X), and shall refer to Lβ(X) as the strong dual of Cp(X). A family {Aα:αNN} of subsets of a set X is called a resolution for X if it covers X and verifies that AαAβ for αβ. If the sets Aα are compact, we speak of a compact resolution. If E is a locally convex space and {Aα:αNN} is a resolution for E consisting of bounded sets (bounded in the locally convex sense [5,1.4.5 Definition]), we say that {Aα:αNN} is a bounded resolution. A bounded resolution for E that swallows the bounded sets of E will be referred to as a fundamental bounded resolution. Recall that a Banach disk B in a locally convex space E is an absolutely convex bounded set such that EB, the linear span of B equipped with the Minkowski functional of B as a norm, is a Banach space. A locally convex space E is said to be a quasi-(LB) -space (in the sense of Valdivia) if it admits a resolution {Aα:αNN} consisting of Banach disks [33].

    If Y is a topological subspace of X, the locally convex space Cp(Y|X) is defined in [2,0.4] as the range (the image) R(T) of Cp(X) in Cp(Y) under the restriction map T:C(X)C(Y), given by ff|Y, equipped with the relative topology of Cp(Y). Note that T is a linear and continuous map. Clearly, C(Y|X)=R(T) consists of those fC(Y) that admit a continuous extension to the whole of X. Observe that Cp(Y|X) is always a dense linear subspace of Cp(Y), so the topological dual of Cp(Y|X) coincides algebraically with L(Y). However, since the weak* topology σ(L(Y),C(Y)) on L(Y) is stronger than the weak* topology σ(L(Y),C(Y|X)), we shall denote by Lp(Y|X) the space L(Y) equipped with this second topology to distinguish it from Lp(Y), which as we know represents the space L(Y) with the topology σ(L(Y),C(Y)). In what follows we shall denote by M(X) the bidual of Cp(X), i.e., the dual of Lβ(X). Likewise, to be consistent with our notation, we shall represent by M(Y|X) the bidual of Cp(Y|X).

    If Y is dense in X, then the restriction map T:Cp(X)Cp(Y) is continuous and one-to-one, i.e., it is a so-called condensation map from Cp(X) onto Cp(Y|X). In this case, Cp(Y|X) is linearly homeomorphic to the space (C(X),τp(Y)), where τp(Y) denotes the locally convex topology on C(X) of the pointwise convergence on Y. Moreover, the adjoint map T:Lp(Y|X)Lp(X) carries Lp(Y|X) onto a dense linear subspace of Lp(X). If Y is closed in X, then Cp(Y|X) is an open subspace of Cp(Y) [2,0.4.1 Proposition]; consequently, in this case the restriction map T is in addition a quotient map. If Y is C-embedded in X (which happens in particular if Y is a compact set in X or if X is normal and Y closed), then T is onto and hence Cp(Y|X)=Cp(Y).

    We denote by δ(Y) the canonical copy of Y consisting of delta measures, i.e., δ(Y)={δy:yY}. The set δ(Y) looks different when regarded as a subset of Lp(Y) or a subset of Lp(Y|X), since in the former case each δy acts as a continuous linear functional on Cp(Y), and in the latter each δy acts as a continuous linear functional on Cp(Y|X). Of course, in the first case δ(Y) is the standard homeomorphic copy of Y in Lp(Y). When a function fC(Y|X) is regarded as a linear functional on wLp(Y|X), we shall write f,w. So, if w=ni=1aiδyiL(Y) with aiR and yiY for 1in, then we have f,w:=ni=1aif(yi).

    Note that C(Y|X)C(Y)RY, which implies that M(Y|X)M(Y)RY. However, the following property holds.

    Theorem 1.1. [18,Corollary 7] The strong duals of Cp(Y|X) and Cp(Y) coincide, as do their biduals M(Y|X) and M(Y).

    In other words, if we denote by Lβ(Y|X) the dual space L(Y) provided with the strong topology β(L(Y),C(Y|X)) of the dual pair L(Y),C(Y|X), it turns out that Lβ(Y|X)=Lβ(Y). Hence, M(Y|X)=M(Y) and, consequently, the weak* bidual of Cp(Y), i.e., the dual space M(Y) equipped with the relative product topology of RY coincides with the weak* bidual of Cp(Y|X). So, the complete picture is

    C(Y|X)C(Y)M(Y|X)=M(Y)RY.

    This result, along with the next three, will be used along the paper.

    Theorem 1.2. [13,Theorem 28] The weak* bidual of Cp(X) admits a bounded resolution if and only if |X|=0.

    Recall that the envelope E={A(αn):αΣ,nN} of a family {Aα:αΣ} of subsets of a locally convex space E with ΣNN, where

    A(αn)={Aβ:βΣ,β(i)=α(i),1in}

    is called limited if for each αΣ and each absolutely convex neighborhood of the origin U in E there is nN with A(αn)nU. A covering {Aα:αΣ} with ΣNN of a set X is called a Σ-covering of X.

    Theorem 1.3. [10,Lemma 2] Let E be a locally convex space. If E admits a Σ -covering {Aα:αΣ} with ΣNN of limited envelope, then there exists a Lindelöf Σ-space Z such that EZRE.

    In the following result, υX stands for the Hewitt realcompactification of X, as usual.

    Theorem 1.4. [10,Theorem 3] The space Cp(X) admits a Σ-covering of limited envelope if and only if υX is a Lindelöf Σ -space.

    It is worth mentioning that each bounded resolution for E is always a Σ -covering of E of limited envelope [10,Proposition 12]. So, in particular, any compact resolution for E is a Σ-covering of E of limited envelope.

    Although Cp(Y|X) is a continuous linear image of Cp(X) and a dense linear subspace of Cp(Y), the topological properties of X and Y induced by certain covering properties of Cp(Y|X) are not always the same as the topological properties of X and Y induced by analogous covering properties of Cp(X) or Cp(Y). For instance, if Cp(X) is σ-countably compact, then X is finite [31]. But if Cp(Y|X) is σ-countably compact and Y is dense in X, then X is pseudocompact and Y is a P-space [2,I.2.2 Theorem]. More generally, if Cp(X) is σ-bounded relatively sequentially complete, then X is finite [17,Corollary 3.2]; but, if Cp(Y|X) is σ-bounded relatively sequentially complete and Y is dense in X, then X is pseudocompact and Y is a P-space [17,Theorem 3.3].

    Motivated by the research carried out in [17,18] on countable coverings of Cp(Y|X) and in [13,14] on uncountable coverings of Cp(X), we extend our study of the locally convex space Cp(Y|X) for (i) fundamental bounded resolutions, (ii) compact resolutions swallowing the compact sets, (iii) resolutions consisting of Banach disks, and (iv) resolutions made up of convex compact sets swallowing the local null sequences. We refer the reader to [2,24,32] for topological and locally convex notions not defined here.

    First, we prove that if Cp(Y|X) has a fundamental resolution consisting of bounded sets, Y must necessarily be countable, which extends [15,Theorem 3.3 (i)] as well as [16,Theorem 1.9], the latter because Cp(X|βX)=Cbp(X). For the definitions and properties of quasi-Suslin and web-compact spaces that we use below, see [32,I.4.2] and [24,Chapter 4], respectively.

    Theorem 2.1. Let YX. Cp(Y|X) admits a fundamental bounded resolution if and only if Y is countable.

    Proof. Assume that {Aα:αNN} is a fundamental resolution for Cp(Y|X) consisting of absolutely convex bounded sets. Since each Aα is absolutely convex, the bipolar theorem ensures that A00α=¯Aα, closure in RY. Consequently, the fact that {Aα:αNN} swallows all bounded sets in Cp(Y|X) means that

    M(Y|X)={¯Aα:αNN}.

    Since M(Y|X) coincides with M(Y), according to Theorem 1.1, we see that M(Y) admits a bounded resolution, namely {¯Aα:αNN}, where each ¯Aα is a compact set in RY. In this case, Theorem 1.2 tells us that |Y|=0.

    Conversely, if Y is countable then Cp(Y) is metrizable, so does Cp(Y|X). But, as is well-known, every metrizable locally convex space E admits a fundamental bounded resolution, namely the family {Aα:αNN} with

    Aα=i=1α(i)Ui,

    where {Ui:iN} is a decreasing base of absolutely convex neighborhoods of the origin in E. So, Cp(Y|X) has a fundamental bounded resolution.

    Corollary 2.1. Let X be a P-space and let Y be a dense subspace of X. Then Cp(Y|X) admits a bounded resolution if and only if Y is countable and discrete.

    Proof. If X is a P-space and Y is dense in X, according to [18,Theorem 18] the space Cp(Y|X) is sequentially complete, and hence locally complete, i.e., such that every bounded set is contained in a Banach disk [26,39.2]. So, if Cp(Y|X) admits a bounded resolution, then Cp(Y|X) is a quasi-(LB)-space. Hence, by [33,Proposition 22] or [24,Theorem 3.5], there is a resolution for Cp(Y|X) consisting of Banach disks that swallows all Banach disks in Cp(Y|X). Thus, the local completeness of Cp(Y|X) guarantees that Cp(Y|X) has a fundamental bounded resolution, which allows Theorem 2.1 to ensure that Y is countable. As every countable P-space is discrete, we are done. Conversely, if Y is countable, then Cp(Y|X), as a linear subspace of RY, is metrizable. Hence, Cp(Y|X) has a bounded resolution.

    Example 2.1. Let X be the one-point Lindelöfication of a discrete space Y with |Y|1. Since Y is uncountable, the previous corollary prevents Cp(Y|X) to admit a bounded resolution.

    Example 2.2. If Y is a dense P-space of X and Cp(Y|X) admits a bounded resolution, even a countable one consisting of sequentially complete bounded sets, Y need not be countable or discrete. In fact, if Y is a nondiscrete P-space and X:=βY, we claim that Cp(Y|X) admits a countable resolution consisting of sequentially complete bounded sets.

    Proof. Clearly, Cp(Y|X)=Cbp(Y), where Cbp(Y) is the dense linear subspace of Cp(Y) consisting of all bounded functions with relative pointwise topology. If B stands for the closed unit ball of the Banach space (Cb(Y),), then C(Y|X)=Cb(Y)=n=1nB. So, the family {Aα:αNN} with Aα=α(1)B for each αNN is a countable bounded resolution for Cp(Y|X). Since Y is a P-space, Cp(Y) is sequentially complete [6]. Hence, if {fn}n=1 is a Cauchy sequence in Cp(Y|X) contained in B, there exists fC(Y) such that fnf in Cp(Y). But, as |fn(y)|1 for all (n,y)N×Y, we have that |f(y)|1 for all yY. Thus, fB, which shows that B is sequentially complete in Cp(Y|X).

    Theorem 2.2. Let Y be a topological subspace of X. If Cp(Y|X) admits a compact resolution that swallows the compact sets, then the compact sets in Y are finite.

    Proof. First note that the topologies σ(L(Y),C(Y)) and σ(L(Y),C(Y|X)) coincide on each compact subset K of Y, or rather on δ(K), regarded as a subset of L(Y). In other words, that each compact set K in Y is embedded both in Lp(Y) and in Lp(Y|X).

    In fact, if a net {yd:dD} in K and a point yK verify that f,δydf,δy for every fC(Y), it is obvious that f,δydf,δy for every fC(Y|X). Thus, δydδy under σ(L(Y),C(Y|X)). Conversely, if a net {yd:dD} in K and a point yK verify that f,δydf,δy for every fC(Y|X), we claim that g,δydg,δy for every gC(Y). Indeed, if gC(Y), then g|KC(K) and there is hC(X) with h|K=g|K. Since h|YC(Y|X), by assumption h|Y,δydh|Y,δy. Thus, h|K,δydh|K,δy, i.e., g,δydg,δy. Consequently, δydδy under σ(L(Y),C(Y)).

    Let {Aα:αNN} be a resolution for Cp(Y|X) consisting of compact sets that swallows the compact sets in Cp(Y|X). To prove that each compact set in Y is finite, we adapt the first part of the argument of [30,3.7 Theorem] (see also [24,Theorem 9.14]).

    Assume by contradiction that there exists an infinite compact set K in Y. As Cp(Y|X) has a compact resolution, it is a quasi-Suslin space in the sense of Valdivia by [7,Proposition 1], hence it is web-compact in the sense of Orihuela (cf. [28]). Thus, the space Cp(Cp(Y|X)) is angelic by [28,Theorem 3].

    Since K is a compact subspace of Y, it follows from the first part of this proof that K is embedded in Cp(Cp(Y|X)). Hence, K is an infinite Fréchet-Urysohn compact subset of the angelic space Cp(Cp(Y|X)) and consequently there exists a non-trivial sequence {xn}n=1 converging to some xK. If we set S:={xn:nN}{x}, then S is a compact countable set, and hence a metrizable compact space. Thus, there is a linear extender map φ:Cp(S)Cp(Y), i.e., such that φ(f|S)=f for every fC(Y), which embeds Cp(S) as a closed linear subspace of Cp(Y), [4,Proposition 4.1]. But, if f:=φ(g) with gC(S), then fC(Y|X). Indeed, if hC(X) extends g to the whole of X, then

    h|Y=φ(h|S)=φ(g)=f,

    which means that fC(Y|X). Thus, φ can be regarded as a map from Cp(S) into Cp(Y|X). Therefore, Cp(S) embeds in Cp(Y|X) as a closed linear subspace of Cp(Y|X). So, the metrizable space Cp(S) contains a resolution of compact sets, namely the family {AαC(S):αNN}, that swallows the compact sets in Cp(S). By Christensen's theorem [8,Theorem 3.3] (see also [12,Theorem 94]) Cp(S) must be a Polish space. Thus, [2,I.3.3 Corollary] guarantees that the compact set S is discrete, and hence finite, a contradiction.

    For the next lemma, recall that, according to [19,15.14 Corollary], a Tychonoff space X is realcompact if and only if its homeomorphic copy δ(X) in Lp(X) is a complete set.

    Theorem 2.3. Assume that X is realcompact and Y is a closed subspace of X. If Cp(Y|X) has a Σ-covering of limited envelope, then Y is a Lindelöf Σ-space.

    Proof. We need the following auxiliary result.

    Claim 2.1. Under our assumptions, the copy δ(Y) of Y in Lp(Y|X) is a complete set.

    Proof of the claim. Let {δyd:dD} be a Cauchy net in the canonical copy δ(Y) of Y in Lp(Y|X), i.e., such that for each fC(Y|X) and ϵ>0 there is d(f,ϵ)D with

    |f(yr)f(ys)|=|f,δyrδys|<ϵ

    for r,sd(f,ϵ). As for each fC(X), one has that f|YC(Y|X), and clearly {δyd:dD}, regarded as a net in δ(X), is a Cauchy net in Lp(X). Since X is realcompact there is zX such that δydδz in Lp(X), i.e., such that f,δydf,δz for every fC(X). But, as Y is closed in X, the canonical copy δ(Y) of Y in Lp(X) is closed in δ(X), which implies that δzδ(Y). This means that zY, and thus we have in particular that f|Y,δydf|Y,δz for every fC(X). From this, it follows that g,δydg,δz for every gC(Y|X), which shows that δ(Y) is a complete set in Lp(Y|X).

    Now, if {Aα:αNN} is a Σ -covering of limited envelope for Cp(Y|X), Theorem 1.3 provides a Lindelöf Σ-space Z such that

    Lp(Y|X)ZRC(Y|X).

    By the claim, δ(Y) is a complete subspace of Lp(Y|X), and consequently, a closed set of the complete locally convex space RC(Y|X). Thus, δ(Y), as a closed topological subspace of Z, is a Lindelöf Σ-space.

    Observe that if {δyd:dD} is a Cauchy net in the copy δ(Y) of Y in Lp(Y|X), then {δyd:dD} need not be a Cauchy net in Lp(Y). So, if we require only Y to be realcompact (regardless of whether X is realcompact or not), the argument of the previous claim does not work.

    Corollary 2.2. Let X be realcompact. If Cp(X) has a bounded resolution, then X has countable extent.

    Proof. If X has uncountable extent, there exists an uncountable closed discrete subspace Y, i.e., consisting of relative isolated points. If {Aα:αNN} is a bounded resolution for Cp(X), the restriction map T:Cp(X)Cp(Y|X) given by T(f)=f|Y maps continuously Cp(X) onto Cp(Y|X). Thus, the family {T(Aα):αNN} is a bounded resolution for Cp(Y|X), and hence a Σ-covering of limited envelope. According to Theorem 2.3, the space Y must be Lindelöf, and hence countable, a contradiction.

    Corollary 2.3. If M denotes the Michael line and P the subspace of the irrational numbers, the following properties hold.

    (1) Cp(M) does not admit a bounded resolution.

    (2) Cp(P|M) does not admit a bounded resolution.

    Proof. Recall that in the Michael line the set P of irrationals is a discrete open subspace. As the topology of M is stronger than that of the real line R and each subspace of R is realcompact, it turns out that M is realcompact by virtue of [19,8.18 Corollary].

    (1) Since there exists an uncountable Euclidean closed set Y in R consisting entirely of irrational numbers, this set Y is closed and discrete in M, so M has uncountable extent and (the contrapositive statement of) Corollary 2.2 applies.

    (2) If Y is again an uncountable closed set in M consisting of irrational numbers, the restriction map S:Cp(P|M)Cp(Y|M) defined as usual by S(f)=f|Y is a continuous linear map. It is also surjective, for if gC(Y|M), there exists fC(M) such that f|Y=g. So, clearly f|PC(P|M) and S(f|P)=f|Y=g. If Cp(P|M) had a bounded resolution, say {Aα:αNN}, then {S(Aα):αNN} would be a bounded resolution for Cp(Y|M). So, according to the proof of Corollary 2.2, the subspace Y would be countable, a contradiction.

    Example 2.3. The converse of Corollary 2.2 does not hold. Let S be the Sorgenfrey line. Since S is a Lindelöf space, it is realcompact and has countable extent. However, Cp(S) does not admit a bounded resolution. Otherwise, by Theorem 1.4, the Sorgenfrey line would be a Lindelöf Σ -space, which is not true since the product of two Lindelöf Σ-spaces is a Lindelöf Σ-space, but S×S is not Lindelöf.

    Corollary 2.4. Let X be realcompact and let Y be a closed subspace which is a P -space. Then Cp(Y|X) admits a bounded resolution if and only if Y is countable and discrete.

    Proof. Since Y is a P-space, each compact set in Y is finite. If, in addition, Cp(Y|X) admits a Σ-covering of limited envelope, then Theorem 2.3 ensures that Y is a Lindelöf Σ-space. But, every Lindelöf Σ-space with finite compact sets is countable (by [2,IV.6.15 Proposition]), and every countable P-space is discrete. The converse is clear.

    Hence, if X is a realcompact P-space and Y a closed subspace, then Cp(Y|X) admits a bounded resolution if and only if Y is countable and discrete.

    Theorem 2.4. Let Y be a closed subspace of a realcompact space X. Then Cp(Y|X) admits a compact resolution that swallows the compact sets if and only if Y is countable and discrete.

    Proof. According to Theorem 2.2, each compact set K in Y is finite. Moreover, as a compact resolution is a Σ-covering of limited envelope, Theorem 2.3 tells us that Y is a Lindelöf Σ -space. But, as we know, a Lindelöf Σ-space with finite compact sets is countable, so Y is countable and, consequently, both spaces Cp(Y|X) and Cp(Y) are metrizable.

    Therefore we have a metrizable topological space Cp(Y|X) with a compact resolution {Aα:αNN} that swallows the compact sets in Cp(Y|X). So, again by Christensen's theorem, Cp(Y|X) is a Polish space, and hence a Čech-complete space [9,4.3.26 Theorem]. Since Cp(Y|X) is a Čech-complete dense subspace of Cp(Y), [2,I.3.1 Theorem] asserts that Y is (countable and) discrete.

    Example 2.4. In the previous theorem, the requirement that X be realcompact is not necessary. In fact, if X is a non-realcompact P -space [19,Problem 9L] and Y is a countable subspace of X, then Y is C-embedded in X, so Cp(Y|X)=Cp(Y)=RY since Y, a countable subspace, is (closed and) discrete. If Y={yn:nN}, then {Aα:αNN} with Aα={fRN:|f(yn)|α(n)} is a compact resolution for Cp(Y|X) that swallows the compact sets.

    Remark 2.1. If X is an arbitrary Tychonoff space and Y=X, using the well-known fact that if {Aα:αNN} is a compact resolution for Cp(X) that swallows the compact sets of Cp(X) and S:Cp(υX)Cp(X) is the restriction map ff|X, then {S1(Aα):αNN} is a compact resolution for Cp(υX) that swallows the compact sets in Cp(υX), and thus, we can deduce [30,3.7 Theorem] from Theorem 2.4. So, if Y=X, we may exempt the space X from being realcompact in the statement of Theorem 2.4.

    Corollary 2.5. Let Y be a closed subspace of a realcompact space X. The following are equivalent

    (1) Cp(Y|X) admits a compact resolution that swallows the compact sets.

    (2) Cp(Y|X) is a Polish space.

    Proof. If Cp(Y|X) has a compact resolution that swallows the compact sets, Y is countable and discrete by Theorem 2.4. Consequently, Cp(Y|X), as a dense linear subspace of Cp(Y)=RY, is a metrizable space with a compact resolution swallowing the compact sets. Hence, a Polish space by Christensen's theorem. Conversely, if Cp(Y|X) is a Polish space, it clearly has a compact resolution swallowing the compact sets.

    Remark 2.2. Let Y be a countable dense subspace of X. If Cp(Y|X) admits a compact resolution that swallows the compact sets, then Y is discrete. In fact, if Yis countable, then Cp(Y|X) is metrizable and we may argue as in the final paragraph of the proof of Theorem 2.4.

    Example 2.5. If Cp(Y|X) admits a bounded resolution that swallows the compact sets in Cp(Y|X), then Y need not be discrete. If M denotes the Michael line, the set Q of rational numbers is a closed subspace of M. Since, RQ=RN clearly has a compact resolution that swallows the compact sets in RQ, we see that Cp(Q|M)=Cp(Q) has a bounded resolution that swallows the compact sets in Cp(Q|M), but Q is not discrete.

    Let us mention that, originally, the definition of quasi-(LB) -spaces emerged as an appropriate range class for the closed graph theorem when strictly barrelled spaces are located at the domain class [33,Corollary 1.5]. Valdivia's class of quasi-(LB) -spaces has proven to be useful both in functional analysis and in topology (see for instance [24,Chapter 3]). Recently, some lifting properties involving quasi-(LB)-spaces at the range class in the closed graph theorem has been relaxed by replacing the strictly barrelled spaces of the domain class by locally convex spaces with a sequential web (see [22,Theorem 1] for details).

    Lemma 3.1. The spaces Cp(X) and Ck(X) have the same Banach disks.

    Proof. If B is a Banach disk in Ck(X), it is clear that B is a Banach disk in Cp(X). Conversely, we claim that if Q is a Banach disk in Cp(X), then Q is a bounded set in Ck(X). Indeed, if U is a basic τk-closed absolutely convex neighborhood of the origin in Ck(X), there is a compact set K in X and ϵ>0 such that

    U={fC(X):supxK|f(x)|ϵ}.

    Clearly, U is a τp-closed set, for if {fd:dD} is a net in U such that fdf in Cp(X), then |f(x)|ϵ for every xK. As the norm topology of EQ is stronger than the pointwise topology, UEQ is also a closed absolutely convex set for the norm topology of EQ. Since E=n=1nU, the Baire category theorem provides some mN such that QmU, which shows that Q is a bounded set in Ck(X), as stated. As Q is a τk-bounded set such that EQ is a Banach space, it turns out that Q is a Banach disk in Ck(X).

    Theorem 3.1. Let X be metrizable. Then Cp(X) is a quasi-(LB)-space if and only if X is σ-compact.

    Proof. If Cp(X) is a quasi-(LB)-space, according to Lemma 3.1 the space Ck(X) is also a quasi-(LB)-space. As X is a kR-space, Ck(X) is complete, and hence locally complete, that is, such that the τk-closed absolutely convex cover of each bounded set in Ck(X) is a Banach disk or, according to [20,Theorem 2.1], such that each convergent sequence in Ck(X) is equicontinuous. Since by [33,Proposition 22] the space Ck(X) admits a resolution consisting of Banach disks that swallows the Banach disks in Ck(X), it follows that Ck(X) has a fundamental bounded resolution. Consequently, by [11,Proposition 3], the space X must be σ-compact.

    Conversely, if X is metrizable and σ-compact, again by [11,Proposition 3] the space Ck(X) has a (fundamental) bounded resolution {Aα:αNN}. Since the local completeness of Ck(X) guarantees that Bα:=¯abx(Aα)τk, i.e. the τk-closed absolutely convex cover of Aα, is a Banach disk for each αNN, and Ck(X) is clearly a quasi-(LB)-space. Thus, Cp(X) is also a quasi-(LB)-space.

    Corollary 3.1. Let Y be a topological subspace of a metrizable space X. If X is σ-compact, then Cp(Y|X) is a quasi-(LB)-space.

    Proof. If X is metrizable and σ-compact, Cp(X) is a quasi-(LB)-space by Theorem 3.1. If T is the restriction map from Cp(X) onto Cp(Y|X), the image T(B) of each Banach disk of Cp(X) is a Banach disk of Cp(Y|X). Thus, Cp(Y|X) is a quasi-(LB)-space.

    Example 3.1. Metrizability cannot be dropped in the 'if' part of Theorem 3.1. If pβNN, equip X=N{p} with the relative topology of βN. Since X is countable, it is σ -compact. However, Cp(X) is not a quasi-(LB)-space. In fact, since Cp(X) is a locally convex Baire space (see [27,Example 7.1]), i.e., such that it cannot be covered by countably many rare, balanced sets [29], if Cp(X) were a quasi-(LB)-space then Cp(X) would be a Fr échet space by virtue of [24,Corollary 3.12]. As Cp(X) is a dense linear subspace of RX, this would imply that Cp(X)=RX. Consequently, the space X should be discrete, which is not true.

    Recall that a sequence {xn}n=1 in a locally convex space E is local null or Mackey convergent to zero [25,28.3] if there is a bounded, closed, absolutely convex set B in E (a closed disk) such that xn0 in the normed space EB:=span(B) equipped with the Minkowski functional of B as a norm. Each local null sequence in E is a null sequence. If E is metrizable, each null sequence is local null [25,28.3.(1) c)]. A linear form u defined on a bornological space E (see [5,3.6.2 Definition] or [25,28]) is continuous if and only if u(xn)0 for each local null sequence {xn}n=1 in E, [25,28.3.(4)]. Recall that, according to the Buchwalter-Schmets theorem [6], the space Cp(X) is bornological if and only if X is realcompact.

    Theorem 4.1. Let Y be a closed subspace of a realcompact space X. If Cp(Y|X) admits a resolution of convex compact sets that swallows the local null sequences in Cp(Y|X), then Y is countable and discrete.

    Proof. As in the proof of [13,Theorem 12], we may assume without loss of generality that Cp(Y|X) admits a resolution {Aα:αNN} of absolutely convex compact sets swallowing the local null sequences in Cp(Y|X).

    We adapt as possible the argument of the proof of [13,Theorem 12] to the present setting. Let M be the family of all local null sequences in Cp(Y|X). Since the family {Aα:αNN} swallows the members of M, the Mackey* topology μ(L(Y),C(Y|X)) of L(Y) is stronger than the topology τc0 on L(Y) of the uniform convergence on the local null sequences of Cp(Y|X). As it is clear that σ(L(Y),C(Y|X))τc0, we conclude that (L(Y),τc0)=C(Y|X). Moreover, as X is realcompact, Cp(X) is bornological, and since Y is a closed subspace of X, the restriction map from Cp(X) onto Cp(Y|X) is a quotient map [2,0.4.1 (2) Proposition], which ensures that Cp(Y|X) is also a bornological space [25,28.4.(2)]. Consequently, the τc0 -dual (L(Y),τc0) of Cp(Y|X) is complete by [25,28.5.(1)]. In fact, it is even μ(L(Y),C(Y|X))-complete by [25,18.4.(4)].

    Now we claim that every compact set in Y is finite. Indeed, if K is a compact set in Y, as we have seen in the proof of Theorem 2.2, the canonical copy δ(K) of K in Lp(Y|X) is embedded in Lp(Y|X), i.e., it is a σ(L(Y),C(Y|X))-compact set in L(Y). So, the completeness of (L(Y),τc0), together with Krein's theorem and the fact that τc0 is a locally convex topology of the dual pair L(Y),C(Y|X), ensures that the weak* closure Q in Lp(Y|X) of the absolutely convex hull of δ(K) is a compact set in Lp(Y|X), and hence a β(L(Y),C(Y|X))-bounded set. On the other hand, as Cp(Y) is a quasibarrelled space [23,11.7.3 Corollary], the β(L(Y),C(Y))-bounded sets in L(Y) are finite-dimensional. But, according to Theorem 1.1, we have that β(L(Y),C(Y))=β(L(Y),C(Y|X)). So, every β(L(Y),C(Y|X))-bounded set in L(Y) is finite-dimensional. Thus, in particular, Q is finite-dimensional. This means that δ(K), as a linearly independent system of vectors in L(Y) contained in Q, must be finite. Hence, the compact set K is finite, as stated.

    Since each compact resolution is a Σ-covering of limited envelope, Theorem 2.3 tells us that Y is a Lindelöf Σ-space. So Y is countable by [2,IV.6.15 Proposition]. Hence, Cp(Y|X) is a metrizable space, which ensures that, in Cp(Y|X), local null sequences and null sequences are the same. Moreover, if M is a compact set in the metrizable space Cp(Y|X), then M lies in the closed absolutely convex cover of a null sequence {fn}n=1, [25,21.10.(3)]. So, if {fn}n=1Aγ, the fact that Aγ is a closed absolutely convex set guarantees that MAγ. Thus, {Aα:αNN} is a compact resolution for Cp(Y|X) that swallows the compact sets of Cp(Y|X). So, we apply Theorem 2.4 to get that Y is (countable and) discrete.

    Remark 4.1. If Y=X, one can get rid of the realcompactness from X by working with Cp(υX) instead of with Cp(X). So, Theorem 4.1 essentially contains [13,Theorem 12].

    Let us examine the existence of a bounded resolution for Cp(Y) that swallows the Cauchy sequences in Cp(Y|X). We denote by R(X) the linear subspace of RX consisting of functions with finite support, i.e., those which vanish off a finite set in X.

    Theorem 5.1. Assume that X is metrizable and Y hemicompact with YX. Then Cp(Y|X) has a bounded resolution that swallows the Cauchy sequences in Cp(Y|X) if and only if Y is countable.

    Proof. We may assume there is a bounded resolution {Aα:αNN} in Cp(Y|X) consisting of absolutely convex sets swallowing the Cauchy sequences in Cp(Y|X). Given eyR(Y) defined by ey(y)=1 and ey(z)=0 if zy, we extend ey to the whole of X by setting ˆey(z)=ey(z) if zY and ˆey(x)=0 if xY. Since X is first countable, as in the proof of [13,Theorem 33] we find a sequence {fy,n}n=1 in C(X) such that fy,nˆey in RX. As fy,n|Yey in Cp(Y), there is γNN with fy,n|YAγ for all nN, so ey¯Aγ. Now, given (n1,α)N×NN, let βNN be such that β(1)=n1 and β(i+1)=α(i) for each iN and define Bβ:=n1¯Aα, closure in RY. Since {ey:yY} is a Hamel basis of R(Y), it follows that {Bβ:βNN} is a bounded family in RY such that R(Y){Bβ:βNN}.

    On the other hand, note that Ck(Y) is a metrizable space due to Arens' theorem [1] (see also [34,Theorem 13.2.4]). As C(Y|X) is clearly a dense linear subspace of Ck(Y), each gCk(Y) is the limit of a sequence {gn}n=1 in C(Y|X) under the compact-open topology τk, and hence under the pointwise topology τp. Thus, there exists δNN such that gnAδ for every nN so that g¯Aδ, which shows that C(Y){Bβ:βNN}. Thus, we conclude that C(Y)+R(Y){Bβ:βNN}, which means that the linear subspace C(Y)+R(Y) of M(Y) admits a bounded resolution under the weak* topology of M(Y). According to [13,Remarks 20,27], Y must be countable.

    For the converse, note that if Y is countable, RY has a compact resolution {Qα:αNN} that swallows the compact sets in RY. Hence, {QαC(Y|X):αNN} is a bounded resolution for Cp(Y|X) such that if {hn}n=1 is a Cauchy sequence in Cp(Y|X) and hRY verifies that hnh in RY, there is γNN with {h,hn:nN}Qγ. Hence, {hn:nN}QγC(Y|X).

    Proposition 5.1. Let Cp(Y) be Fréchet-Urysohn. If Cp(Y|X) has a bounded resolution that swallows the Cauchy sequences in Cp(Y|X), then Cp(Y) has a bounded resolution.

    Proof. Let {Aα:αNN} be a bounded resolution for Cp(Y|X) that swallows the Cauchy sequences in Cp(Y|X). As C(Y)M(Y)=M(Y|X), the latter equality because of Theorem 1.1, given gC(Y) there is a bounded set Q in Cp(Y|X) such that g¯Q, closure in Cp(Y). But, since Cp(Y) is Fré chet-Urysohn, there exists a sequence {gn}n=1 in QC(Y|X) such that gng in Cp(Y). As {gn}n=1 is a Cauchy sequence in Cp(Y|X), there is γNN such that gnAγ for every nN. Consequently, we have g¯Aγ, closure in Cp(Y). Since each bounded set in Cp(Y|X) is a bounded set in Cp(Y) and the closure of a bounded set is bounded, it follows that the family {¯Aα:αNN} is a bounded resolution for Cp(Y).

    Corollary 5.1. Assume that Y is a Lindelöf P-space, with YX. Then Cp(Y|X) has a bounded resolution that swallows the Cauchy sequences in Cp(Y|X) if and only if Y is countable and discrete.

    Proof. Since Y is a Lindelöf P-space, it turns out that Cp(Y) is Fréchet-Urysohn [3,10.2 Theorem]. Consequently, Cp(Y) has a bounded resolution by the preceding proposition. So, according to [15,Proposition 3.6], the subspace Y must be countable and discrete. The proof of the converse is analogous to that of Theorem 5.1.

    If F is a linear subspace of C(X), let Y={xX:δxF} so that δ(Y)=δ(X)F. Since F, the annihilator of F in L(X), is a closed subspace of Lp(X), Y is an open subset of X. Define T:FCp(Y|X) by the rule Tf=f|Y.

    Theorem 6.1. T is a linear homeomorphism from F onto its range, which embeds F isomorphically into Cp(Y|X).

    Proof. Let us first check that T is one-to-one. So, if fkerTf|Y=0, we have to show that f=0. Equivalently, if fF is such that f0, we must prove that Tf0. But, if f0, there is uL(X) with f,u0. If u=v+w with v=pi=1aiδxi, where δxiF or, equivalently, xiXY for 1ip, and w=qi=1aiδyi with δyiF or, equivalently, yiY for 1iq, then

    0f,u=f,w=f|Y,w=Tf,w,

    since wL(Y). As L(Y) is the dual of Cp(Y|X), this shows that Tf0, as desired.

    Let us check that T1:ImTF is continuous when ImT is provided with the relative topology of Cp(Y|X) and F is equipped with the relative topology of Cp(X). Indeed, if {fd:dD} is a net in F and fF are such that TfdTf in ImT, it is clear that fd(y)f(y) for every yY. Now, if xX, either δxF or δxF. In the first case, it is obvious that fd(x)=00=f(x) since fd,fF. In the second case, xY, so fd(x)f(x). This shows that fdf in F pointwise on X. Since T is continuous, one-to-one, and a topological homomorphism, it turns out that T is a linear homeomorphism that embeds isomorphically F into Cp(Y|X).

    Theorem 6.2. Let Z be a closed set in X. If F={fC(X):f(z)=0 zZ}, then F is linearly homeomorphic to a dense subspace of Cp(Y|X).

    Proof. Since it is clear that δzF for each zZ, we have ZXY. But, if xXY and xZ, there is gC(X) with g(x)=1 and g(z)=0 for all zZ, which means that gF. But g,δx=1, so δxF, i.e., xY, a contradiction. Therefore, XY=Z. Thus, if T:FCp(Y|X) is the map considered in the previous theorem, then

    T(F)={gC(Y):fC(X),f|Y=g,f|XY=0}.

    By Theorem 6.1 the subspace T(F) of Cp(Y|X) is isomorphic to F. If Δ is a finite subset of Y and hC(Y|X), the fact that Z is closed allows us to find fC(X) such that f(y)=h(y) for each yΔ and f(z)=0 for every zZ, i.e., a function fC(X) such that g:=f|Y lies in T(F) and verifies that g(y)=h(y) for yΔ. This shows that T(F) is dense in Cp(Y|X).

    Example 6.1. If X=D{ξ} is the one-point Lindelöfication of the discrete space D, the subspace Σ(D) of RD consisting of all countably supported functions on D is linearly homeomorphic to a dense subspace of Cp(D|X). In particular, Cp(D|X) is a Baire space.

    Proof. If F:={fC(X):f(ξ)=0}, the restriction map S:FΣ(D) given by Sf=f|D is a linear homeomorphism from F onto Σ(D). If we put Y:={xX:δxF}, Theorem 6.2 asserts that the subspace F of Cp(X) is isomorphic to a dense subspace of Cp(Y|X). Consequently, Σ(D) is isomorphic to a dense subspace of Cp(Y|X).

    But, clearly D=Y, since, if xD, there is fC(X) with f(x)=1 and f(ξ)=0, such that fF and f,δx=1, i.e., δxF. Thus, xY. Whereas if xY, then δxF, which means that xξ, so xD. The second statement follows from the fact that Σ(D) is a Baire space.

    Remark 6.1. If X is uncountable, then Σ(X) does not admit a bounded resolution. This is because Σ(X) is locally complete. Thus, Valdivia's theorem [24,Theorem 3.5] ensures that Σ(X) admits a bounded resolution if and only if it is a quasi-(LB)-space. Since Σ(X) is a Baire space, Σ(X) is necessarily a Fréchet space [24,Corollary 3.12], which implies that Σ(X)=RX with X countable. Note that this implies that if X=D{ξ} is the one-point Lindelöfication of the discrete space D with |D|1, then Cp(D|X) does not have a bounded resolution (cf. Example 2.1).

    Example 6.2. If X=D{ξ} is the one-point compactification of the discrete space D, the subspace R(D) of RD consisting of all finitely supported functions on D is linearly homeomorphic to a dense subspace of the space Cp(D|X).

    Proof. If F:={fC(X):f(ξ)=0}, the linear map S:FR(D) defined by Sf=f|D is a linear homeomorphism from the closed one-codimensional linear subspace F of Cp(X) onto R(D). Likewise, the set Y:={xX:δxF} coincides with D since δ(X)F={δξ}. By Theorem 6.2, F is linearly homeomorphic to a dense subspace of Cp(Y|X). So, R(D) is linearly isomorphic to a dense linear subspace of Cp(D|X).

    Remark 6.2. R(X) has a fundamental bounded resolution if and only if X is countable.

    Proof. Let {Aα:αNN} be a fundamental resolution for R(X) consisting of absolutely convex bounded sets. Thus, the bipolar theorem ensures that A00α=¯Aα, closure in RX. Hence, the fact that {Aα:αNN} swallows all compact sets in R(X) means that the bidual E of R(X) is given by E={¯Aα:αNN}. As each ¯Aα is an absolutely convex compact set in RX, and thus a Banach disk in E, it turns out that E is a quasi-(LB)-space. On the other hand, since each fΣ(X) is the limit of a sequence in R(X), we have that Σ(X)E. As Σ(X) is a dense Baire subspace of RX, it follows that E is a Baire space. Since each locally convex space which is both a quasi-(LB)-space and a Baire space is a Fréchet space, necessarily E=RX with X countable. The converse is obvious.

    As a consequence, if X=Y{ξ} is the one-point compactification of the discrete space Y, then Cp(Y|X) admits a fundamental bounded resolution if and only if Y is countable, which is a particular case of Theorem 2.1.

    Juan C. Ferrando: conceptualization, research, methodology, formal analysis, writing-original draft, review & editing, validation; Manuel López-Pellicer: formal analysis, validation, writing review; Santiago Moll-López: validation, writing-review & editing. All authors have read and agreed to the published version of the manuscript.

    The authors declare they have not used Artificial Intelligence (AI) tools in the creation of this article.

    The first named author is supported by Generalitat Valenciana under project PROMETEO/2021/063.

    The authors declare that they have no conflict of interest.



    [1] J. Q. Gong, J. Jin, A better robustness and fast convergence zeroing neural network for solving dynamic nonlinear equations, Neural Comput. Appl., 2021, 1–11. doi: 10.1007/s00521-020-05617-9.
    [2] L. Xie, J. Ding, F. Ding, Gradient based iterative solutions for general linear matrix equations, Comput. Math. Appl., 58 (2009), 1441–1448. doi: 10.1016/j.camwa.2009.06.047.
    [3] Z. N. Zhang, F. Ding, X. G. Liu, Hierarchical gradient based iterative parameter estimation algorithm for multivariable output error moving average systems, Comput. Math. Appl., 61 (2011), 672–682. doi: 10.1016/j.camwa.2010.12.014.
    [4] F. Ding, Combined state and least squares parameter estimation algorithms for dynamic systems, Appl. Math. Modell., 38 (2014), 403–412. doi: 10.1016/j.apm.2013.06.007.
    [5] H. M. Zhang, Quasi gradient-based inversion-free iterative algorithm for solving a class of the nonlinear matrix equations, Comput. Math. Appl., 77 (2019), 1233–1244. doi: 10.1016/j.camwa.2018.11.006.
    [6] H. M. Zhang, L. J. Wan, Zeroing neural network methods for solving the Yang-Baxter-like matrix equation, Neurocomputing, 383 (2020), 409–418. doi: 10.1016/j.neucom.2019.11.101.
    [7] V. N. Katsikis, S. D. Mourtas, P. S. Stanimirovicˊ, Y. N. Zhang, Solving complex-valued time-varying linear matrix equations via QR decomposition with applications to robotic motion tracking and on angle-of-arrival localization, IEEE T. Neur. Net. Lear., 2021, 1–10. doi: 10.1109/TNNLS.2021.3052896.
    [8] L. Xiao, Z. J. Zhang, S. Li, Solving time-varying system of nonlinear equations by finite-time recurrent neural networks with application to motion tracking of robot manipulators, IEEE T. Syst. Man Cy.-S., 49 (2019), 2210–2220. doi: 10.1109/TSMC.2018.2836968.
    [9] L. Xiao, S. Li, K. L. Li, L. Jin, B. L. Liao, Co-design of finite-time convergence and noise suppression: A unified neural model for time varying linear equations with robotic applications, IEEE T. Syst. Man Cy.-S., 50 (2020), 5233–5243. doi: 10.1109/TSMC.2018.2870489.
    [10] L. Jin, S. Li, B. L. Liao, Z. J. Zhang, Zeroing neural networks: A survey, Neurocomputing, 267 (2017), 597–604. doi: 10.1016/j.neucom.2017.06.030.
    [11] Z. B. Sun, T. Shi, L. Jin, B. C. Zhang, Z. X. Pang, J. Z. Yu, Discrete-time zeroing neural network of O(\tau^4) pattern for online time-varying nonlinear optimization problem: Application to manipulator motion generation, J. Frankl. I., 358 (2021), 7203–7220. doi: 10.1016/j.jfranklin.2021.07.006.
    [12] L. Jin, Y. N. Zhang, Discrete-time Zhang neural network for online time-varying nonlinear optimization with application to manipulattor motion generation, IEEE T. Neur. Net. Lear., 26 (2015), 1525–1531. doi: 10.1109/TNNLS.2014.2342260.
    [13] L. Jin, S. Li, Distributed task allocation of multiple robots: A control perspective, IEEE T. Syst. Man Cy.-S., 48 (2018), 693–701. doi: 10.1109/TSMC.2016.2627579.
    [14] Y. M. Qi, L. Jin, Y. N. Wang, L. Xiao, J. L. Zhang, Complex-valued discrete-time neural dynamics for perturbed time-dependent complex quadratic programming with applications, IEEE T. Neur. Net. Lear., 31 (2020), 3555–3569. doi: 10.1109/TNNLS.2019.2944992.
    [15] L. Jin, Y. N. Zhang, Continuous and discrete Zhang dynamics for real-time varying nonlinear optimization, Numer. Algorithms, 73 (2016), 115–140. doi: 10.1007/s11075-015-0088-1.
    [16] L. Jin, Y. N. Zhang, S. Li, Y. Y. Zhang, Noise-tolerant ZNN models for solving time-varying zero-finding problems: A control-theoretic approach, IEEE T. Automat. Contr., 62 (2017), 992–997. doi: 10.1109/TAC.2016.2566880.
    [17] Z. B. Sun, T. Shi, L. Wei, Y. Y. Sun, K. P. Liu, L. Jin, Noise-suppressing zeroing neural network for online solving time-varying nonlinear optimization problem: A control-based approach, Neural Comput. Appl., 32 (2020), 11505–11520. doi: 10.1007/s00521-019-04639-2.
    [18] Z. B. Sun, F. Li, B. C. Zhang, Y. Y. Sun, L. Jin, Different modified zeroing neural dynamics with inherent tolerance to noises for time-varying reciprocal problems: A control-theoretic approach, Neurocomputing, 337 (2019), 165–179. doi: 10.1016/j.neucom.2019.01.064.
    [19] S. Z. Qiao, X. Z. Wang, Y. M. Wei, Two finite-time convergent Zhang neural network models for time-varying complex matrix Drazin inverse, Linear Algebra Appl., 542 (2018), 101–117. doi: 10.1016/j.laa.2017.03.014.
    [20] Z. Li, Y. N. Zhang, Improved Zhang neural network model and its solution of time-varying generalized linear matrix equations, Expert Syst. Appl., 37 (2010), 7213–7218. doi: 10.1016/j.eswa.2010.04.007.
    [21] S. Li, S. F. Chen, B. Liu, Accelerating a recurrent neural network to finite-time convergence for solving time-varying Sylvester equation by using a sign-bi-power activation function, Neural Process. Lett., 37 (2013), 189–205. doi: 10.1007/s11063-012-9241-1.
    [22] Y. J. Shen, P. Miao, Y. H. Huang, Y. Shen, Finite-time stability and its application for solving time-varying Sylvester equation by recurrent neural network, Neural Process. Lett., 42 (2015), 763–784. doi: 10.1007/s11063-014-9397-y.
    [23] L. Xiao, Accelerating a recurrent neural network to finite-time convergence using a new design formula and its application to time-varying matrix square root, J. Frankl. I., 354 (2017), 5667–5677. doi: 10.1016/j.jfranklin.2017.06.012.
    [24] L. Jia, L. Xiao, J. H. Dai, Z. H. Qi, Z. J. Zhang, Y. S. Zhang, Design and application of an adaptive fuzzy control strategy to zeroing neural network for solving time-variant QP problem, IEEE T. Fuzzy Syst., 29 (2021), 1544–1555. doi: 10.1109/TFUZZ.2020.2981001.
    [25] V. N. Katsikis, S. D. Mourtas, P. S. Stanimirovi\rm\acute{c}, Y. N. Zhang, Continuous-time varying complex QR decomposition via zeroing neural dynamics, Neural Process. Lett., 53 (2021), 3573–3590. doi: 10.1007/s11063-021-10566-y.
    [26] Z. J. Zhang, L. N. Zheng, J. Weng, Y. J. Mao, W. Lu, L. Xiao, A new varying-parameter recurrent neural-network for online solution of time-varying Sylvester equation, IEEE T. Cybernetics, 48 (2018), 3135–3148. doi: 10.1109/TCYB.2017.2760883.
    [27] Z. J. Zhang, L. N. Zheng, T. R. Qiu, F. Q. Deng, Varying-parameter convergent-differential neural solution to time-varying overdetermined system of linear equations, IEEE T. Automat. Contr., 65 (2020), 874–881. doi: 10.1109/TAC.2019.2921681.
    [28] Z. J. Zhang, L. N. Zheng, A complex varying-parameter convergent-differential neural-network for solving online time-varying complex Sylvester equation, IEEE T. Cybernetics, 49 (2019), 3627–3639. doi: 10.1109/TCYB.2018.2841970.
    [29] Z. J. Zhang, L. D. Kong, L. N. Zheng, P. C. Zhang, X. L. Qu, B. L. Liao, et al., Robustness analysis of a power-type varying-parameter recurrent neural network for solving time-varying QM and QP problems and applications, IEEE T. Syst. Man Cy.-S., 50 (2020), 5106–5118. doi: 10.1109/TSMC.2018.2866843.
    [30] H. M. Zhang, F. Ding, On the Kronecker products and their applications, J. Appl. Math., 2013 (2013), 1–8. doi: 10.1155/2013/296185.
    [31] Y. N. Zhang, D. C. Jiang, J. Wang, A recurrent neural network for solving Sylvester equation with time-varying coefficients, IEEE T. Neural Networ., 13 (2002), 1053–1063 doi: 10.1109/TNN.2002.1031938.
    [32] K. Chen, Recurrent implicit dynamics for online matrix inversion, Appl. Math. Comput., 219 (2013), 10218–10224. doi: 10.1016/j.amc.2013.03.117.
    [33] Y. N. Zhang, K. Chen, H. Z. Tan, Performance analysis of gradient neural network exploited for online time-varying matrix inversion, IEEE T. Automat. Contr., 54 (2009), 1940–1945. doi: 10.1109/TAC.2009.2023779.
    [34] F. Ding, G. J. Liu, X. P. Liu, Parameter estimation with scarce measurements, Automatica, 47 (2011), 1646–1655. doi: 10.1016/j.automatica.2011.05.007.
    [35] F. Ding, Y. J. Liu, B. Bao, Gradient based and least squares based iterative estimation algorithms for multi-input multi-output systems, P. I. Mech. Eng. I-J. Sys., 226 (2012), 43–55. doi: 10.1177/0959651811409491.
    [36] F. Ding, G. J. Liu, X. P. Liu, Partially coupled stochastic gradient identification methods for non-uniformly sampled systems, IEEE T. Automat. Contr., 55 (2010), 1976–1981. doi: 10.1109/TAC.2010.2050713.
    [37] L. Xiao, J. H. Dai, L. Jin, W. B. Li, S. Li, J. Hou, A noise-enduring and finite-time zeroing neural network for equality-constrained time-varying nonlinear optimization, IEEE T. Syst. Man Cy.-S., 51 (2021), 4729–4740. doi: 10.1109/TSMC.2019.2944152.
    [38] L. Xiao, K. L. Li, M. X. Duan, Computing time-varying quadratic optimization with finite-time convergence and noise tolerance: A unified framework for zeroing neural network, IEEE Trans. Neural Netw. Lear. Syst., 30 (2019), 3360–3369. doi: 10.1109/TNNLS.2019.2891252.
    [39] L. Xiao, Y. S. Zhang, J. H. Dai, K. Chen, S. Yang, W. B. Li, et al., A new noise-tolerant and predefined-time ZNN model for time-dependent matrix inversion, Neural Networks, 117 (2019), 124–134. doi: 10.1016/j.neunet.2019.05.005.
    [40] F. Yu, L. Liu, L. Xiao, K. L. Li, S. Cai, A robust and fixed-time zeroing neural dynamics for computing time-variant nonlinear equation using a novel nonliear activation function, Neurocomputing, 350 (2019), 108–116. doi: 10.1016/j.neucom.2019.03.053.
    [41] L. Xiao, Y. K. Cao, J. H. Dai, L. Jia, H. Y. Tan, Finite-time and predefined-time convergence design for zeroing neural network: Theorem, method, and verification, IEEE T. Ind. Inform., 17 (2021), 4724–4732. doi: 10.1109/TII.2020.3021438.
    [42] L. Xiao, J. H. Dai, R. B. Lu, S. Li, J. C. Li, S. J. Wang, Design and comprehensive analysis of a noise-tolerant ZNN model with limited-time convergence for time-dependent nonlinear minimization, IEEE T. Neur. Net. Lear., 31 (2020), 5339–5348. doi: 10.1109/TNNLS.2020.2966294.
    [43] L. Xiao, Y. S. Zhang, Q. Y. Zuo, J. H. Dai, J. C. Li, W. S. Tang, A noise-tolerant zeroing neural network for time-dependent complex matrix inversion under various kinds of noises, IEEE T. Ind. Inform., 16 (2020), 3757–3766. doi: 10.1109/TII.2019.2936877.
    [44] M. Liu, L. M. Chen, X. H. Du, L. Jin, M. S. Shang, Activated gradients for deep neural networks, IEEE T. Neur. Net. Lear., 2021, 1–13. doi: 10.1109/TNNLS.2021.3106044.
  • Reader Comments
  • © 2022 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(2196) PDF downloads(93) Cited by(3)

Figures and Tables

Figures(5)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog