Control of Semilinear Differential Equations with Moving Singularities

Abstract

In this paper, we present a control problem related to a semilinear differential equation with a moving singularity, i.e., the singular point depends on a parameter. The particularity of the controllability condition resides in the fact that it depends on the singular point, which in turn depends on the control variable. We provide sufficient conditions to ensure that the functional determining the control is continuous over the entire domain of the parameter. Lower and upper solutions techniques combined with a bisection algorithm is used to prove the controllability of the equation and to approximate the control. An example is given together with some numerical simulations. The results naturally extend to fractional differential equations.

Authors

Radu Precup
Faculty of Mathematics and Computer Science and Institute of Advanced Studies in Science and Technology, Babes-Bolyai, University, Cluj-Napoca, Romania
Tiberiu Popoviciu Institute of Numerical Analysis, Romanian Academy, Romania

Andrei Stan
Tiberiu Popoviciu Institute of Numerical Analysis, Romanian Academy, Romania
Department of Mathematics, Babes-Bolyai University, Cluj-Napoca, Romania

Wei-Shih Du
Department of Mathematics, National Kaohsiung Normal University, Kaohsiung 824004, Taiwan.

Keywords

 

control problemmoving singularitydifferential equationlower and upper solutionsbisection algorithmfractional differential equationnumerical simulation

Paper coordinates

R. Precup, A. Stan, W.-S. Du, Control of Semilinear Differential Equations with Moving Singularities, Fractal and Fractional, 9 (2025) no. 4, art. no. 198. https://doi.org/10.3390/fractalfract9040198

About this paper

Journal
Axioms

Fractal and Fractional

Publisher Name

MDPI

Print ISSN
Online ISSN

google scholar link

References

[1] Torres, P.J. Mathematical Models with Singularities: A Zoo of Singular Creatures; Atlantis Press: Amsterdam, The Netherlands, 2015. [Google Scholar]
[2] O’Regan, D. Theory of Singular Boundary Value Problems; World Scientific Publishing: River Edge, NJ, USA, 1994. [Google Scholar]
[3] Callegari, A.; Nachman, A. A nonlinear singular boundary value problem in the theory of pseudoplasitc fluids. SIAM J. Appl. Math. 1980, 38, 275–282. [Google Scholar]
[4] Gingold, H. Rosenblat, S. Differential equations with moving singularities. SIAM J. Math. Anal 1976, 7, 942–957. [Google Scholar]
[5] Gingold, H. Introduction to differential equations with moving singularities. Rocky Mt. J. Math. 1976, 6, 571–574. [Google Scholar] [CrossRef]
[6] Fila, M.; Takahashi, J.; Yanagida, E. Solutions with moving singularities for a one-dimensional nonlinear diffusion equation. Math. Ann. 2024, 390, 5383–5413. [Google Scholar] [CrossRef]
[7] Nicolescu, M.; Dinculeanu, N.; Marcus, S. Manual de Analiză Matematicăa (II); Editura Didactică şi Pedagogică: Bucureşti, Romania, 1964. (In Romanian) [Google Scholar]
[8] Haplea, I.Ş.; Parajdi, L.G.; Precup, R. On the controllability of a system modeling cell dynamics related to leukemia. Symmetry 2021, 13, 1867. [Google Scholar] [CrossRef]
[9] Precup, R. On some applications of the controllability principle for fixed point equations. Results Appl. Math. 2022, 13, 100236. [Google Scholar] [CrossRef]
[10] Coron, J.M. Control and Nonlinearity; AMS: Providence, RI, USA, 2007. [Google Scholar]
[11] Coddington, E.A. An Introduction to Ordinary Differential Equations; Dover: New York, NY, USA, 1961. [Google Scholar]
[12] Precup, R. Ordinary Differential Equations; De Gruyter: Berlin, Germany, 2018. [Google Scholar]
[13] O’Regan, D.; Precup, R. Theorems of Leray-Schauder Type and Applications; CRC Press: Boca Raton, FL, USA, 2002. [Google Scholar]
[14] Ciarlet, G. Linear and Nonlinear Functional Analysis with Applications; SIAM: Philadelphia, PA, USA, 2013. [Google Scholar]
[15] Le Dret, H. Nonlinear Elliptic Partial Differential Equations; Springer: Berlin, Germany, 2018. [Google Scholar]
[16] Gronwall, T.H. Note on the derivatives with respect to a parameter of the solutions of a system of differential equations. Ann. Math. 1919, 20, 292–296. [Google Scholar]
[17] Pachpatte, B.G.; Ames, W.F. Inequalities for Differential and Integral Equations; Academic Press: Cambridge, MA, USA, 1997. [Google Scholar]
[18] Rudin, W. Principles of Mathematical Analysis; MacGraw-Hill: New York, NY, USA, 1976. [Google Scholar]
[19] Parajdi, L.G.; Precup, R.; Haplea, I.S. A method of lower and upper solutions for control problems and application to a model of bone marrow transplantation. Int. J. Appl. Math. Comput. Sci. 2023, 33, 409–418. [Google Scholar]
[20] Samko, S.G.; Kilbas, A.A.; Marichev, O.I. Fractional Integrals and Derivatives: Theory and Applications; Gordon and Breach Science Publishers: Yverdon, Switzerland, 1993. [Google Scholar]
[21] Kilbas, A.A.; Trujillo, J.J. Differentiale equations of fractional order: Methods, results and problems-I. Appl. Anal. 2001, 78, 153–192. [Google Scholar]
[22] Benchohra, M.; Hamani, S.; Ntouyas, S.K. Boundary value problems for differential equations with fractional order and nonlocal conditions. Nonlinear Anal. 2009, 71, 2391–2396. [Google Scholar] [CrossRef]
[23] Podlubny, I. Fractional Differential Equations; Mathematics in Sciences and Engineering; Academic Press: San Diego, CA, USA, 1999; Volume 198. [Google Scholar]

Paper (preprint) in HTML form

Control of semilinear differential equations with moving singularities

Radu Precup R. Precup, Faculty of Mathematics and Computer Science and Institute of Advanced Studies in Science and Technology, Babeş-Bolyai University, 400084 Cluj-Napoca, Romania & Tiberiu Popoviciu Institute of Numerical Analysis, Romanian Academy, P.O. Box 68-1, 400110 Cluj-Napoca, Romania r.precup@ictp.acad.ro Andrei Stan A. Stan, Department of Mathematics, Babeş-Bolyai University, 400084 Cluj-Napoca, Romania & Tiberiu Popoviciu Institute of Numerical Analysis, Romanian Academy, P.O. Box 68-1, 400110 Cluj-Napoca, Romania andrei.stan@ubbcluj.ro  and  Wei-Shih Du W-S. Du, Department of Mathematics, National Kaohsiung Normal University, Kaohsiung 82444, Taiwan wsdu@mail.nknu.edu.tw
Abstract.

In this paper, we present a control problem related to a semilinear differential equation with a moving singularity, i.e., the singular point depends on a parameter. The particularity of the controllability condition resides in the fact that it depends on the singular point which in turn depends on the control variable. We provide sufficient conditions to ensure that the functional determining the control is continuous over the entire domain of the parameter. Lower and upper solutions technique combined with a bisection algorithm is used to prove the controllability of the equation and to approximate the control. An example is given together with some numerical simulations. The results naturally extend to fractional differential equations.

1. Introduction and preliminaries

Differential equations are crucial in solving practical problems in many scientific fields, such as physics, chemistry, biology, economics, and engineering, etc., modeling many real-world processes. However, the complexity of these phenomena often introduces various parameters that can significantly influence the outcome. A particularly intriguing problem in this context is identifying the parameters that ensure a specific quantity (e.g., density, energy) related to the solution of the differential equation reaches a desired value. This challenge naturally leads to a control problem.

Our study has two strong motivations.

(a):

The first motivation concerns differential equations with moving singularities, which frequently appear in nonlinear models from applied sciences, such as physics and mathematical biology [1].

(b):

The second one relates to the control of such models, aiming to reach a desired state of the process. For example, if the state variable represents a density, one might be interested in controlling its cumulative value or average. This corresponds precisely to our control problem in Section 2.

Mathematical models expressed through equations with singularities include the Briot-Bouquet equation, which has applications in complex analysis, specifically in the theory of univalent functions; equations arising in Michaelis-Menten kinetics, modeling oxygen diffusion in cells; the Thomas-Fermi equation in atomic physics; and the Emden-Fowler equation, in the study of phenomena in non-Newtonian fluid mechanics [1, 2, 3].

Inspired by the investigation in [4, 5, 6], this paper will explore the following problem

(1.1) {u(t)=f(t,u(t),λ),t[0,θ(λ))u(0)=u0(λ).casesformulae-sequencesuperscript𝑢𝑡𝑓𝑡𝑢𝑡𝜆𝑡0𝜃𝜆otherwise𝑢0subscript𝑢0𝜆otherwise\begin{cases}u^{\prime}(t)=f(t,u(t),\lambda),\quad t\in\left[0,\,\theta(% \lambda)\right)\\ u(0)=u_{0}(\lambda).\end{cases}{ start_ROW start_CELL italic_u start_POSTSUPERSCRIPT ′ end_POSTSUPERSCRIPT ( italic_t ) = italic_f ( italic_t , italic_u ( italic_t ) , italic_λ ) , italic_t ∈ [ 0 , italic_θ ( italic_λ ) ) end_CELL start_CELL end_CELL end_ROW start_ROW start_CELL italic_u ( 0 ) = italic_u start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT ( italic_λ ) . end_CELL start_CELL end_CELL end_ROW

Here, f:[0,)××(0,)¯(:={±}):𝑓00annotated¯assignabsentplus-or-minus\ f:[0,\infty)\times\mathbb{R}\times(0,\infty)\rightarrow\bar{\mathbb{R}}(:=% \mathbb{R\cup\{\pm\infty\}})italic_f : [ 0 , ∞ ) × blackboard_R × ( 0 , ∞ ) → over¯ start_ARG blackboard_R end_ARG ( := blackboard_R ∪ { ± ∞ } ) is a function that possesses a singularity in the first variable, influenced by the third one, that is, for each λ>0𝜆0\lambda>0italic_λ > 0, there exists θ(λ)>0𝜃𝜆0\theta(\lambda)>0italic_θ ( italic_λ ) > 0 such that

limtθ(λ)t<θ(λ)f(t,x,λ)=±for almost all x.formulae-sequencesubscript𝑡𝜃𝜆𝑡𝜃𝜆𝑓𝑡𝑥𝜆plus-or-minusfor almost all 𝑥\lim_{\begin{subarray}{c}t\rightarrow\,\theta(\lambda)\\ t<\,\theta(\lambda)\end{subarray}}f(t,x,\lambda)=\pm\infty~{}\quad\text{for % almost all }x\in\mathbb{R}.roman_lim start_POSTSUBSCRIPT start_ARG start_ROW start_CELL italic_t → italic_θ ( italic_λ ) end_CELL end_ROW start_ROW start_CELL italic_t < italic_θ ( italic_λ ) end_CELL end_ROW end_ARG end_POSTSUBSCRIPT italic_f ( italic_t , italic_x , italic_λ ) = ± ∞ for almost all italic_x ∈ blackboard_R .

Throughout this paper, we use uλsubscript𝑢𝜆u_{\lambda}italic_u start_POSTSUBSCRIPT italic_λ end_POSTSUBSCRIPT to denote the unique solution to problem (1.1) for a given λ>0𝜆0\lambda>0italic_λ > 0. Since the singularity point θ(λ)𝜃𝜆\theta(\lambda)italic_θ ( italic_λ ) varies with λ𝜆\lambdaitalic_λ, the differential equation (1.1) is said to be with moving singularity.

For each λ>0𝜆0\lambda>0italic_λ > 0, we consider the functional ψλ:C[0,θ(λ))¯:subscript𝜓𝜆𝐶0𝜃𝜆¯\psi_{\lambda}\colon C[0,\theta(\lambda))\rightarrow\bar{\mathbb{R}}italic_ψ start_POSTSUBSCRIPT italic_λ end_POSTSUBSCRIPT : italic_C [ 0 , italic_θ ( italic_λ ) ) → over¯ start_ARG blackboard_R end_ARG,

(1.2) ψλ(u)=0θ(λ)u(s)𝑑s,subscript𝜓𝜆𝑢superscriptsubscript0𝜃𝜆𝑢𝑠differential-d𝑠\psi_{\lambda}(u)=\int_{0}^{\theta(\lambda)}u(s)\,ds,italic_ψ start_POSTSUBSCRIPT italic_λ end_POSTSUBSCRIPT ( italic_u ) = ∫ start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_θ ( italic_λ ) end_POSTSUPERSCRIPT italic_u ( italic_s ) italic_d italic_s ,

where the integration over the noncompact interval [0,θ(λ))0𝜃𝜆\left[0,\,\theta(\lambda)\right)[ 0 , italic_θ ( italic_λ ) ) is understood in the usual sense (see, e.g., [7]),

0θ(λ)u(s)𝑑s=limtθ(λ)t<θ(λ)0tu(s)𝑑s.superscriptsubscript0𝜃𝜆𝑢𝑠differential-d𝑠subscript𝑡𝜃𝜆𝑡𝜃𝜆superscriptsubscript0𝑡𝑢𝑠differential-d𝑠\int_{0}^{\theta(\lambda)}u(s)\,ds=\lim_{\begin{subarray}{c}t\rightarrow\theta% (\lambda)\\ t<\theta(\lambda)\end{subarray}}\int_{0}^{t}u(s)\,ds.∫ start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_θ ( italic_λ ) end_POSTSUPERSCRIPT italic_u ( italic_s ) italic_d italic_s = roman_lim start_POSTSUBSCRIPT start_ARG start_ROW start_CELL italic_t → italic_θ ( italic_λ ) end_CELL end_ROW start_ROW start_CELL italic_t < italic_θ ( italic_λ ) end_CELL end_ROW end_ARG end_POSTSUBSCRIPT ∫ start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_t end_POSTSUPERSCRIPT italic_u ( italic_s ) italic_d italic_s .

Our goal in this paper is the following control problem:

Problem (control problem).

Find λ>0superscript𝜆0\lambda^{\ast}>0italic_λ start_POSTSUPERSCRIPT ∗ end_POSTSUPERSCRIPT > 0 such that

(1.3) ψλ(uλ)=p,subscript𝜓superscript𝜆subscript𝑢superscript𝜆𝑝\psi_{\lambda^{\ast}}(u_{\lambda^{\ast}})=p,italic_ψ start_POSTSUBSCRIPT italic_λ start_POSTSUPERSCRIPT ∗ end_POSTSUPERSCRIPT end_POSTSUBSCRIPT ( italic_u start_POSTSUBSCRIPT italic_λ start_POSTSUPERSCRIPT ∗ end_POSTSUPERSCRIPT end_POSTSUBSCRIPT ) = italic_p ,

where p𝑝p\in\mathbb{R}italic_p ∈ blackboard_R is a given value.

The particularity of the controllability condition resides in the fact that it depends on the singular point which in turn depends on the control variable. To establish sufficient conditions for the existence of a solution to this control problem, we first guarantee that the mapping φ:(0,):𝜑0\varphi\colon\left(0,\infty\right)\rightarrow\mathbb{R}italic_φ : ( 0 , ∞ ) → blackboard_R,

(1.4) φ(λ)=ψλ(uλ),𝜑𝜆subscript𝜓𝜆subscript𝑢𝜆\varphi(\lambda)=\psi_{\lambda}(u_{\lambda}),italic_φ ( italic_λ ) = italic_ψ start_POSTSUBSCRIPT italic_λ end_POSTSUBSCRIPT ( italic_u start_POSTSUBSCRIPT italic_λ end_POSTSUBSCRIPT ) ,

is well-defined and continuous on (0,)0(0,\infty)( 0 , ∞ ). Then, we are able to use a lower and upper solution argument to guarantee the existence of a λsuperscript𝜆\lambda^{\ast}italic_λ start_POSTSUPERSCRIPT ∗ end_POSTSUPERSCRIPT with the desired property (1.3). Moreover, by bisection algorithm, we have a method of approximation of the value λ.superscript𝜆\lambda^{\ast}.italic_λ start_POSTSUPERSCRIPT ∗ end_POSTSUPERSCRIPT .

Control problems on a fixed interval that require to find a parameter in order to achieve a specific controllability condition are well-documented in the literature (see, e.g., [8, 9, 10]). The novelty of this paper lies in determining such a parameter where each solution is defined on a different interval. This approach requires a more refined analysis and leads to more complex problems.

The main assumptions we use in our analysis are

  • (h1)

    For each λ>0𝜆0\lambda>0italic_λ > 0 and 0<ε<θ(λ)0𝜀𝜃𝜆0<\varepsilon<\theta(\lambda)0 < italic_ε < italic_θ ( italic_λ ), there exists a constant L=L(λ,ε)𝐿𝐿𝜆𝜀L=L(\lambda,\varepsilon)italic_L = italic_L ( italic_λ , italic_ε ) such that for all x,x¯𝑥¯𝑥x,\overline{x}\in\mathbb{R}italic_x , over¯ start_ARG italic_x end_ARG ∈ blackboard_R and for all t[0,θ(λ)ε]𝑡0𝜃𝜆𝜀t\in\left[0,\,\theta(\lambda)-\varepsilon\right]italic_t ∈ [ 0 , italic_θ ( italic_λ ) - italic_ε ], we have

    |f(t,x,λ)f(t,x¯,λ)|L(λ,ε)|xx¯|.𝑓𝑡𝑥𝜆𝑓𝑡¯𝑥𝜆𝐿𝜆𝜀𝑥¯𝑥|f(t,x,\lambda)-f(t,\overline{x},\lambda)|\leq L(\lambda,\varepsilon)\,|x-% \overline{x}|.| italic_f ( italic_t , italic_x , italic_λ ) - italic_f ( italic_t , over¯ start_ARG italic_x end_ARG , italic_λ ) | ≤ italic_L ( italic_λ , italic_ε ) | italic_x - over¯ start_ARG italic_x end_ARG | .
  • (h2)

    The mappings

    λθ(λ),λu0(λ),λL(λ,ε)(for each ε>0),formulae-sequencemaps-to𝜆𝜃𝜆formulae-sequencemaps-to𝜆subscript𝑢0𝜆maps-to𝜆𝐿𝜆𝜀for each 𝜀0\lambda\mapsto\theta(\lambda),\quad\lambda\mapsto u_{0}(\lambda),\quad\lambda% \mapsto L(\lambda,\varepsilon)\,\,(\text{for each }\varepsilon>0),italic_λ ↦ italic_θ ( italic_λ ) , italic_λ ↦ italic_u start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT ( italic_λ ) , italic_λ ↦ italic_L ( italic_λ , italic_ε ) ( for each italic_ε > 0 ) ,

    are continuous. Additionally, the map (t,x,λ)f(t,x,λ)maps-to𝑡𝑥𝜆𝑓𝑡𝑥𝜆(t,x,\lambda)\mapsto f(t,x,\lambda)( italic_t , italic_x , italic_λ ) ↦ italic_f ( italic_t , italic_x , italic_λ ) is continuous for t[0,θ(λ))𝑡0𝜃𝜆t\in[0,\,\theta(\lambda))italic_t ∈ [ 0 , italic_θ ( italic_λ ) ), x𝑥x\in\mathbb{R}italic_x ∈ blackboard_R, and λ(0,)𝜆0\lambda\in(0,\infty)italic_λ ∈ ( 0 , ∞ ).

In the next lemma, we show that assumptions (h1), (h2) are sufficient to guarantee the existence of a unique solution of problem (1.1).

Lemma 1.1.

Under assumptions (h1) and (h2), for each λ>0𝜆0\lambda>0italic_λ > 0, there exists a unique solution uλC[0,θ(λ))subscript𝑢𝜆𝐶0𝜃𝜆u_{\lambda}\in C[0,\,\theta(\lambda))italic_u start_POSTSUBSCRIPT italic_λ end_POSTSUBSCRIPT ∈ italic_C [ 0 , italic_θ ( italic_λ ) ) of problem (1.1)))). Moreover, this solution satisfies the integral equation

(1.5) uλ(t)=u0(λ)+0tf(s,uλ(s),λ)𝑑s,subscript𝑢𝜆𝑡subscript𝑢0𝜆superscriptsubscript0𝑡𝑓𝑠subscript𝑢𝜆𝑠𝜆differential-d𝑠u_{\lambda}(t)=u_{0}(\lambda)+\int_{0}^{t}f(s,u_{\lambda}(s),\lambda)\,ds,italic_u start_POSTSUBSCRIPT italic_λ end_POSTSUBSCRIPT ( italic_t ) = italic_u start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT ( italic_λ ) + ∫ start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_t end_POSTSUPERSCRIPT italic_f ( italic_s , italic_u start_POSTSUBSCRIPT italic_λ end_POSTSUBSCRIPT ( italic_s ) , italic_λ ) italic_d italic_s ,

for all t[0,θ(λ)).𝑡0𝜃𝜆t\in\left[0,\,\theta(\lambda)\right).italic_t ∈ [ 0 , italic_θ ( italic_λ ) ) .

Proof.

Let λ>0𝜆0\lambda>0italic_λ > 0. For each 0<ε<θ(λ)0𝜀𝜃𝜆0<\varepsilon<\theta(\lambda)0 < italic_ε < italic_θ ( italic_λ ), we consider the initial value problem on the cut-off domain,

(1.6) {u(t)=f(t,u(t),λ),t[0,θ(λ)ε]u(0)=u0(λ).casessuperscript𝑢𝑡𝑓𝑡𝑢𝑡𝜆𝑡0𝜃𝜆𝜀𝑢0subscript𝑢0𝜆otherwise\begin{cases}u^{\prime}(t)=f(t,u(t),\lambda),&t\in\left[0,\,\theta(\lambda)-% \varepsilon\right]\\ u(0)=u_{0}(\lambda).&\end{cases}{ start_ROW start_CELL italic_u start_POSTSUPERSCRIPT ′ end_POSTSUPERSCRIPT ( italic_t ) = italic_f ( italic_t , italic_u ( italic_t ) , italic_λ ) , end_CELL start_CELL italic_t ∈ [ 0 , italic_θ ( italic_λ ) - italic_ε ] end_CELL end_ROW start_ROW start_CELL italic_u ( 0 ) = italic_u start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT ( italic_λ ) . end_CELL start_CELL end_CELL end_ROW

From assumption (h1), the mapping xf(t,x,λ)maps-to𝑥𝑓𝑡𝑥𝜆x\mapsto f(t,x,\lambda)italic_x ↦ italic_f ( italic_t , italic_x , italic_λ ) is Lipschitz continuous with the Lipschitz constant Lλ,εsubscript𝐿𝜆𝜀L_{\lambda,\varepsilon}italic_L start_POSTSUBSCRIPT italic_λ , italic_ε end_POSTSUBSCRIPT. Therefore, problem (1.6) has a unique solution (see, e.g., [12, 11]). The conclusion follows immediately by letting ε0𝜀0\varepsilon\rightarrow 0italic_ε → 0 and using the uniqueness of the solution on each interval [0,θ(λ)ε]0𝜃𝜆𝜀[0,\theta(\lambda)-\varepsilon][ 0 , italic_θ ( italic_λ ) - italic_ε ]. Relation (1.5) can be easily deduced by taking the integral in (1.1) from 00 to t.𝑡t.italic_t .

Example A.

The model for f(t,x,λ)𝑓𝑡𝑥𝜆f(t,x,\lambda)italic_f ( italic_t , italic_x , italic_λ ) is given by

f(t,x,λ)=xλt.𝑓𝑡𝑥𝜆𝑥𝜆𝑡f(t,x,\lambda)=\frac{x}{\lambda-t}.italic_f ( italic_t , italic_x , italic_λ ) = divide start_ARG italic_x end_ARG start_ARG italic_λ - italic_t end_ARG .

Clearly, there is a singularity at θ(λ)=λ𝜃𝜆𝜆\theta(\lambda)=\lambdaitalic_θ ( italic_λ ) = italic_λ. However, when the first variable is restricted to a compact interval [0,λε]0𝜆𝜀[0,\lambda-\varepsilon][ 0 , italic_λ - italic_ε ] with 0<ε<θ(λ)0𝜀𝜃𝜆0<\varepsilon<\theta(\lambda)0 < italic_ε < italic_θ ( italic_λ ), the function f𝑓fitalic_f is Lipschitz continuous with L(λ,ε)=1ε𝐿𝜆𝜀1𝜀L(\lambda,\varepsilon)=\frac{1}{\varepsilon}italic_L ( italic_λ , italic_ε ) = divide start_ARG 1 end_ARG start_ARG italic_ε end_ARG the Lipschitz constant. If we set u0(λ)=1λ>0subscript𝑢0𝜆1𝜆0u_{0}(\lambda)=\frac{1}{\lambda}>0italic_u start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT ( italic_λ ) = divide start_ARG 1 end_ARG start_ARG italic_λ end_ARG > 0, the unique solution of the problem (1.1) is

uλ(t)=1λt,for all t[0,λ).formulae-sequencesubscript𝑢𝜆𝑡1𝜆𝑡for all 𝑡0𝜆u_{\lambda}(t)=\frac{1}{\lambda-t}\,,\ \ \text{for all }t\in[0,\lambda).italic_u start_POSTSUBSCRIPT italic_λ end_POSTSUBSCRIPT ( italic_t ) = divide start_ARG 1 end_ARG start_ARG italic_λ - italic_t end_ARG , for all italic_t ∈ [ 0 , italic_λ ) .
Remark 1.1.

Without imposing further assumptions beyond (h1) and (h2), we cannot generally expect φ𝜑\varphiitalic_φ to be well-defined and continuous on (0,)0(0,\infty)( 0 , ∞ ). For instance, consider the case where

f(t,x,λ)=1λ(λt)1λ+1andu0(λ)=1λ1/λ.formulae-sequence𝑓𝑡𝑥𝜆1𝜆superscript𝜆𝑡1𝜆1andsubscript𝑢0𝜆1superscript𝜆1𝜆f(t,x,\lambda)=\frac{1}{\lambda(\lambda-t)^{\frac{1}{\lambda}+1}}\quad\text{% and}\quad u_{0}(\lambda)=\frac{1}{\lambda^{1/\lambda}}.italic_f ( italic_t , italic_x , italic_λ ) = divide start_ARG 1 end_ARG start_ARG italic_λ ( italic_λ - italic_t ) start_POSTSUPERSCRIPT divide start_ARG 1 end_ARG start_ARG italic_λ end_ARG + 1 end_POSTSUPERSCRIPT end_ARG and italic_u start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT ( italic_λ ) = divide start_ARG 1 end_ARG start_ARG italic_λ start_POSTSUPERSCRIPT 1 / italic_λ end_POSTSUPERSCRIPT end_ARG .

Clearly, both assumptions (h1) and (h2) are satisfied. Straightforward calculations yield

uλ(t)=(λt)1λ,subscript𝑢𝜆𝑡superscript𝜆𝑡1𝜆u_{\lambda}(t)=(\lambda-t)^{-\frac{1}{\lambda}},italic_u start_POSTSUBSCRIPT italic_λ end_POSTSUBSCRIPT ( italic_t ) = ( italic_λ - italic_t ) start_POSTSUPERSCRIPT - divide start_ARG 1 end_ARG start_ARG italic_λ end_ARG end_POSTSUPERSCRIPT ,

for each λ>1𝜆1\lambda>1italic_λ > 1. Consequently,

φ(λ)=1λ1λ2λ1λ.𝜑𝜆1𝜆1superscript𝜆2𝜆1𝜆\varphi(\lambda)=\frac{1}{\lambda-1}\lambda^{\frac{2\lambda-1}{\lambda}}.italic_φ ( italic_λ ) = divide start_ARG 1 end_ARG start_ARG italic_λ - 1 end_ARG italic_λ start_POSTSUPERSCRIPT divide start_ARG 2 italic_λ - 1 end_ARG start_ARG italic_λ end_ARG end_POSTSUPERSCRIPT .

This implies that φ(λ)𝜑𝜆\varphi(\lambda)\in\mathbb{R}italic_φ ( italic_λ ) ∈ blackboard_R for λ>1𝜆1\lambda>1italic_λ > 1, but

limλ1λ>1φ(λ)=.subscript𝜆1𝜆1𝜑𝜆\lim_{\begin{subarray}{c}\lambda\rightarrow 1\\ \lambda>1\end{subarray}}\varphi(\lambda)=\infty.roman_lim start_POSTSUBSCRIPT start_ARG start_ROW start_CELL italic_λ → 1 end_CELL end_ROW start_ROW start_CELL italic_λ > 1 end_CELL end_ROW end_ARG end_POSTSUBSCRIPT italic_φ ( italic_λ ) = ∞ .

The paper is structured as follows: Section 2 presents the original results on the controllability of equations with moving singularities. Before establishing the main result (Theorem 2.6), we derive several auxiliary results concerning the continuous dependence of solutions on the control parameter and the continuity of the control functional with respect to the control variable. In Section 3, we provide a theoretical algorithm for obtaining the solution of the control problem using the method of lower and upper solutions. We also provide an example that illustrates the applicability of the obtained results. Finally, in Section 4 we suggest a possible extension of our approach to fractional differential equations with moving singularities.

In the following, we present some well-known results from the literature that will be used throughout this paper. The first result is the Arzelà-Ascoli theorem (see, e.g., [13, 14]).

Theorem 1.2 (Arzelà-Ascoli theorem).

A subset F(C[a,b],||)F\subset\left(C[a,b],|\cdot|_{\infty}\right)italic_F ⊂ ( italic_C [ italic_a , italic_b ] , | ⋅ | start_POSTSUBSCRIPT ∞ end_POSTSUBSCRIPT ), where |||\cdot|_{\infty}| ⋅ | start_POSTSUBSCRIPT ∞ end_POSTSUBSCRIPT is the supremum norm, is relatively compact if and only if it is uniformly bounded and equicontinuous, that is, there exists M>0𝑀0M>0italic_M > 0 with

|u|M,for all uF,formulae-sequencesubscript𝑢𝑀for all 𝑢𝐹|u|_{\infty}\leq M,\quad\text{for all }u\in F,| italic_u | start_POSTSUBSCRIPT ∞ end_POSTSUBSCRIPT ≤ italic_M , for all italic_u ∈ italic_F ,

and for every ε>0𝜀0\varepsilon>0italic_ε > 0, there exists δ(ε)>0𝛿𝜀0\delta(\varepsilon)>0italic_δ ( italic_ε ) > 0 such that

|u(x)u(y)|ε,for x,y[a,b] with |xy|δ(ε) and all uF.formulae-sequence𝑢𝑥𝑢𝑦𝜀for 𝑥𝑦𝑎𝑏 with 𝑥𝑦𝛿𝜀 and all 𝑢𝐹\left|u(x)-u(y)\right|\leq\varepsilon,\quad\text{for }x,y\in[a,b]\text{ with }% \left|x-y\right|\leq\delta(\varepsilon)\text{ and all }u\in F.| italic_u ( italic_x ) - italic_u ( italic_y ) | ≤ italic_ε , for italic_x , italic_y ∈ [ italic_a , italic_b ] with | italic_x - italic_y | ≤ italic_δ ( italic_ε ) and all italic_u ∈ italic_F .

The next lemma provides an alternative condition to ensure the convergence of a sequence based on the behavior of its subsequences (see, e.g., [15, Lemma 1.1]).

Theorem 1.3.

Let X𝑋Xitalic_X be a topological space, and let (xn)subscript𝑥𝑛(x_{n})( italic_x start_POSTSUBSCRIPT italic_n end_POSTSUBSCRIPT ) be a sequence in X𝑋Xitalic_X with the following property: there exists xX𝑥𝑋x\in Xitalic_x ∈ italic_X such that from any subsequence of (xn)subscript𝑥𝑛(x_{n})( italic_x start_POSTSUBSCRIPT italic_n end_POSTSUBSCRIPT ), a further subsequence can be extracted that converges to x𝑥xitalic_x. Then the entire sequence (xn)subscript𝑥𝑛(x_{n})( italic_x start_POSTSUBSCRIPT italic_n end_POSTSUBSCRIPT ) converges to x𝑥xitalic_x.

Finally, we have the well known Grönwall’s lemma (see, [16, 17]).

Theorem 1.4 (Grönwall inequality).

Let u,vC[a,b]𝑢𝑣𝐶𝑎𝑏u,v\in C[a,b]italic_u , italic_v ∈ italic_C [ italic_a , italic_b ]. If there exists a constant c>0𝑐0c>0italic_c > 0 such that

|u(t)|c+at|u(s)||v(s)|𝑑sfor t[a,b],formulae-sequence𝑢𝑡𝑐superscriptsubscript𝑎𝑡𝑢𝑠𝑣𝑠differential-d𝑠for 𝑡𝑎𝑏\left|u(t)\right|\leq c+\int_{a}^{t}|u(s)|\,|v(s)|\,ds~{}\quad\text{for }t\in[% a,b],| italic_u ( italic_t ) | ≤ italic_c + ∫ start_POSTSUBSCRIPT italic_a end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_t end_POSTSUPERSCRIPT | italic_u ( italic_s ) | | italic_v ( italic_s ) | italic_d italic_s for italic_t ∈ [ italic_a , italic_b ] ,

then

|u(t)|cexp(at|v(s)|𝑑s)for t[a,b].formulae-sequence𝑢𝑡𝑐superscriptsubscript𝑎𝑡𝑣𝑠differential-d𝑠for 𝑡𝑎𝑏\left|u(t)\right|\leq c\exp\left(\int_{a}^{t}|v(s)|\,ds\right)~{}\quad\text{% for }t\in[a,b].| italic_u ( italic_t ) | ≤ italic_c roman_exp ( ∫ start_POSTSUBSCRIPT italic_a end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_t end_POSTSUPERSCRIPT | italic_v ( italic_s ) | italic_d italic_s ) for italic_t ∈ [ italic_a , italic_b ] .

3. Approximate solving of the control problem

Starting from the lower and upper solutions λ¯,¯𝜆\underline{\lambda},under¯ start_ARG italic_λ end_ARG , λ¯,¯𝜆\overline{\lambda},over¯ start_ARG italic_λ end_ARG , one can approximate λsuperscript𝜆\lambda^{\ast}italic_λ start_POSTSUPERSCRIPT ∗ end_POSTSUPERSCRIPT by using the following algorithm. For a similar use of this method, we refer the reader to [19].

Algorithm (Bisection algorithm).


Step 0 (initialization): k:=0,λ¯0:=λ¯,formulae-sequenceassign𝑘0assignsubscript¯𝜆0¯𝜆k:=0,\ \underline{\lambda}_{0}:=\underline{\lambda},italic_k := 0 , under¯ start_ARG italic_λ end_ARG start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT := under¯ start_ARG italic_λ end_ARG , λ¯0:=assignsubscript¯𝜆0absent\overline{\lambda}_{0}:=over¯ start_ARG italic_λ end_ARG start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT := λ¯¯𝜆\overline{\lambda}over¯ start_ARG italic_λ end_ARG.

Step k(k1)::𝑘𝑘1absentk\ \left(k\geq 1\right):italic_k ( italic_k ≥ 1 ) : compute λ:=λ¯k1+λ¯k12;assign𝜆subscript¯𝜆𝑘1subscript¯𝜆𝑘12\lambda:=\frac{\underline{\lambda}_{k-1}+\overline{\lambda}_{k-1}}{2};italic_λ := divide start_ARG under¯ start_ARG italic_λ end_ARG start_POSTSUBSCRIPT italic_k - 1 end_POSTSUBSCRIPT + over¯ start_ARG italic_λ end_ARG start_POSTSUBSCRIPT italic_k - 1 end_POSTSUBSCRIPT end_ARG start_ARG 2 end_ARG ;

  • \bullet

    If φ(λ)=p,𝜑𝜆𝑝\varphi\left(\lambda\right)=p,italic_φ ( italic_λ ) = italic_p , then λ=λsuperscript𝜆𝜆\lambda^{\ast}=\lambdaitalic_λ start_POSTSUPERSCRIPT ∗ end_POSTSUPERSCRIPT = italic_λ and we are finished;

  • \bullet

    If φ(λ)<p,𝜑𝜆𝑝\varphi\left(\lambda\right)<p,italic_φ ( italic_λ ) < italic_p , then set λ¯k:=λassignsubscript¯𝜆𝑘𝜆\underline{\lambda}_{k}:=\lambdaunder¯ start_ARG italic_λ end_ARG start_POSTSUBSCRIPT italic_k end_POSTSUBSCRIPT := italic_λ and λ¯k:=λ¯k1assignsubscript¯𝜆𝑘subscript¯𝜆𝑘1\overline{\lambda}_{k}:=\overline{\lambda}_{k-1}over¯ start_ARG italic_λ end_ARG start_POSTSUBSCRIPT italic_k end_POSTSUBSCRIPT := over¯ start_ARG italic_λ end_ARG start_POSTSUBSCRIPT italic_k - 1 end_POSTSUBSCRIPT and repeat Step k𝑘kitalic_k with k:=k+1;assign𝑘𝑘1k:=k+1;italic_k := italic_k + 1 ;

  • \bullet

    If φ(λ)>p,𝜑𝜆𝑝\varphi\left(\lambda\right)>p,italic_φ ( italic_λ ) > italic_p , then set λ¯k:=λ¯k1assignsubscript¯𝜆𝑘subscript¯𝜆𝑘1\underline{\lambda}_{k}:=\underline{\lambda}_{k-1}under¯ start_ARG italic_λ end_ARG start_POSTSUBSCRIPT italic_k end_POSTSUBSCRIPT := under¯ start_ARG italic_λ end_ARG start_POSTSUBSCRIPT italic_k - 1 end_POSTSUBSCRIPT and λ¯k:=λassignsubscript¯𝜆𝑘𝜆\overline{\lambda}_{k}:=\lambdaover¯ start_ARG italic_λ end_ARG start_POSTSUBSCRIPT italic_k end_POSTSUBSCRIPT := italic_λ and repeat Step k𝑘kitalic_k with k:=k+1;assign𝑘𝑘1k:=k+1;italic_k := italic_k + 1 ;

Stop criterion: if |φ(λ)p|ε,𝜑𝜆𝑝𝜀\left|\varphi\left(\lambda\right)-p\right|\leq\varepsilon,| italic_φ ( italic_λ ) - italic_p | ≤ italic_ε , then λλsimilar-to-or-equals𝜆superscript𝜆\lambda\simeq\lambda^{\ast}italic_λ ≃ italic_λ start_POSTSUPERSCRIPT ∗ end_POSTSUPERSCRIPT (with error ε𝜀\varepsilonitalic_ε).

We note that this step-by-step algorithm iteratively approximates the control solution. At each step, based on the obtained feedback, either the subsolution or the supersolution is improved.

Theorem 3.1.

Under assumptions (h1)-(h3), the bisection algorithm is convergent to a solution λsuperscript𝜆\lambda^{\ast}italic_λ start_POSTSUPERSCRIPT ∗ end_POSTSUPERSCRIPT of the control problem.

Proof.

If the algorithm does not stop after a finite number of steps, then it generates two bounded and monotone (so convergent) sequences (λ¯k)subscript¯𝜆𝑘\left(\underline{\lambda}_{k}\right)( under¯ start_ARG italic_λ end_ARG start_POSTSUBSCRIPT italic_k end_POSTSUBSCRIPT ) and (λ¯k),subscript¯𝜆𝑘\left(\overline{\lambda}_{k}\right),( over¯ start_ARG italic_λ end_ARG start_POSTSUBSCRIPT italic_k end_POSTSUBSCRIPT ) , which in addition satisfy

(3.1) |λ¯kλ¯k|=|λ¯λ¯|2k(k0),subscript¯𝜆𝑘subscript¯𝜆𝑘¯𝜆¯𝜆superscript2𝑘𝑘0\left|\overline{\lambda}_{k}-\underline{\lambda}_{k}\right|=\frac{\left|% \overline{\lambda}-\underline{\lambda}\right|}{2^{k}}\ \ \left(k\geq 0\right),| over¯ start_ARG italic_λ end_ARG start_POSTSUBSCRIPT italic_k end_POSTSUBSCRIPT - under¯ start_ARG italic_λ end_ARG start_POSTSUBSCRIPT italic_k end_POSTSUBSCRIPT | = divide start_ARG | over¯ start_ARG italic_λ end_ARG - under¯ start_ARG italic_λ end_ARG | end_ARG start_ARG 2 start_POSTSUPERSCRIPT italic_k end_POSTSUPERSCRIPT end_ARG ( italic_k ≥ 0 ) ,
(3.2) φ(λ¯k)<p,φ(λ¯k)>p.formulae-sequence𝜑subscript¯𝜆𝑘𝑝𝜑subscript¯𝜆𝑘𝑝\varphi\left(\underline{\lambda}_{k}\right)<p,\ \ \ \varphi\left(\overline{% \lambda}_{k}\right)>p.italic_φ ( under¯ start_ARG italic_λ end_ARG start_POSTSUBSCRIPT italic_k end_POSTSUBSCRIPT ) < italic_p , italic_φ ( over¯ start_ARG italic_λ end_ARG start_POSTSUBSCRIPT italic_k end_POSTSUBSCRIPT ) > italic_p .

From (3.1), the two sequences have the same limit denoted λ,superscript𝜆\lambda^{\ast},italic_λ start_POSTSUPERSCRIPT ∗ end_POSTSUPERSCRIPT , while from (3.2), in virtue of the continuity Theorem 2.5, we obtain

φ(λ)pand φ(λ)p.formulae-sequence𝜑superscript𝜆𝑝and 𝜑superscript𝜆𝑝\varphi\left(\lambda^{\ast}\right)\leq p\ \ \ \text{and\ \ \ }\varphi\left(% \lambda^{\ast}\right)\geq p.italic_φ ( italic_λ start_POSTSUPERSCRIPT ∗ end_POSTSUPERSCRIPT ) ≤ italic_p and italic_φ ( italic_λ start_POSTSUPERSCRIPT ∗ end_POSTSUPERSCRIPT ) ≥ italic_p .

Hence φ(λ)=p𝜑superscript𝜆𝑝\varphi\left(\lambda^{\ast}\right)=pitalic_φ ( italic_λ start_POSTSUPERSCRIPT ∗ end_POSTSUPERSCRIPT ) = italic_p as desired. ∎

Example B.

A typical example of function f𝑓fitalic_f satisfying conditions (h1)-(h3) is

(3.3) f(t,x,λ)=xa(λt),𝑓𝑡𝑥𝜆𝑥𝑎𝜆𝑡f(t,x,\lambda)=\frac{x}{a(\lambda-t)},italic_f ( italic_t , italic_x , italic_λ ) = divide start_ARG italic_x end_ARG start_ARG italic_a ( italic_λ - italic_t ) end_ARG ,

where a>1𝑎1a>1italic_a > 1. Clearly, θ(λ)=λ𝜃𝜆𝜆\theta(\lambda)=\lambdaitalic_θ ( italic_λ ) = italic_λ. If in addition we take u0(λ)=λ1asubscript𝑢0𝜆superscript𝜆1𝑎u_{0}(\lambda)=\lambda^{-\frac{1}{a}}italic_u start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT ( italic_λ ) = italic_λ start_POSTSUPERSCRIPT - divide start_ARG 1 end_ARG start_ARG italic_a end_ARG end_POSTSUPERSCRIPT, we obtain the unique solution of problem (1.1),

uλ(t)=1(λt)1a.subscript𝑢𝜆𝑡1superscript𝜆𝑡1𝑎u_{\lambda}(t)=\frac{1}{(\lambda-t)^{\frac{1}{a}}}.italic_u start_POSTSUBSCRIPT italic_λ end_POSTSUBSCRIPT ( italic_t ) = divide start_ARG 1 end_ARG start_ARG ( italic_λ - italic_t ) start_POSTSUPERSCRIPT divide start_ARG 1 end_ARG start_ARG italic_a end_ARG end_POSTSUPERSCRIPT end_ARG .

Therefore,

φ(λ)=0λuλ(s)𝑑s=aa1λa1a,𝜑𝜆superscriptsubscript0𝜆subscript𝑢𝜆𝑠differential-d𝑠𝑎𝑎1superscript𝜆𝑎1𝑎\varphi(\lambda)=\int_{0}^{\lambda}u_{\lambda}(s)\,ds=\frac{a}{a-1}\lambda^{% \frac{a-1}{a}},italic_φ ( italic_λ ) = ∫ start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_λ end_POSTSUPERSCRIPT italic_u start_POSTSUBSCRIPT italic_λ end_POSTSUBSCRIPT ( italic_s ) italic_d italic_s = divide start_ARG italic_a end_ARG start_ARG italic_a - 1 end_ARG italic_λ start_POSTSUPERSCRIPT divide start_ARG italic_a - 1 end_ARG start_ARG italic_a end_ARG end_POSTSUPERSCRIPT ,

which is well-defined and continuous on (0,)0(0,\infty)( 0 , ∞ ).

Refer to caption
Figure 1. Error decay in the bisection method.

We conclude this example with some numerical simulations for the function f𝑓fitalic_f defined in (3.3), with a=3𝑎3a=3italic_a = 3. Our aim is to determine λsuperscript𝜆\lambda^{\ast}italic_λ start_POSTSUPERSCRIPT ∗ end_POSTSUPERSCRIPT such that φ(λ)=3𝜑superscript𝜆3\varphi(\lambda^{\ast})=3italic_φ ( italic_λ start_POSTSUPERSCRIPT ∗ end_POSTSUPERSCRIPT ) = 3. The exact value is known to be λexact=232subscript𝜆exactsuperscript232\lambda_{\text{exact}}=2^{\frac{3}{2}}italic_λ start_POSTSUBSCRIPT exact end_POSTSUBSCRIPT = 2 start_POSTSUPERSCRIPT divide start_ARG 3 end_ARG start_ARG 2 end_ARG end_POSTSUPERSCRIPT. For the lower and upper solutions of the control problem, we take λ¯=1¯𝜆1\underline{\lambda}=1under¯ start_ARG italic_λ end_ARG = 1 and λ¯=4¯𝜆4\overline{\lambda}=4over¯ start_ARG italic_λ end_ARG = 4, respectively, while the tolerance ε𝜀\varepsilonitalic_ε is chosen to be ε=106𝜀superscript106\varepsilon=10^{-6}italic_ε = 10 start_POSTSUPERSCRIPT - 6 end_POSTSUPERSCRIPT.

In the Figure 1, the blue curve represents the error between φ(λ)𝜑𝜆\varphi(\lambda)italic_φ ( italic_λ ) and p=3𝑝3p=3italic_p = 3, where λ𝜆\lambdaitalic_λ at each step k𝑘kitalic_k is λ:=λ¯k1+λ¯k12,assign𝜆subscript¯𝜆𝑘1subscript¯𝜆𝑘12\lambda:=\frac{\underline{\lambda}_{k-1}+\overline{\lambda}_{k-1}}{2},italic_λ := divide start_ARG under¯ start_ARG italic_λ end_ARG start_POSTSUBSCRIPT italic_k - 1 end_POSTSUBSCRIPT + over¯ start_ARG italic_λ end_ARG start_POSTSUBSCRIPT italic_k - 1 end_POSTSUBSCRIPT end_ARG start_ARG 2 end_ARG , while the orange curve represents the difference between the calculated value of λ𝜆\lambdaitalic_λ and the exact value λexactsubscript𝜆exact\lambda_{\text{exact}}italic_λ start_POSTSUBSCRIPT exact end_POSTSUBSCRIPT. After 18 iterations, the approximate value of the control is found to be

λ=2.828848838806.superscript𝜆2.828848838806\lambda^{\ast}=2.828848838806.italic_λ start_POSTSUPERSCRIPT ∗ end_POSTSUPERSCRIPT = 2.828848838806 .

In Figure 2, the graph of uλsubscript𝑢𝜆u_{\lambda}italic_u start_POSTSUBSCRIPT italic_λ end_POSTSUBSCRIPT is plotted for the last three values of λ𝜆\lambdaitalic_λ obtained from the bisection algorithm (those corresponding to the lowest error in the previous figure). We see that for λ=2.82𝜆2.82\lambda=2.82italic_λ = 2.82, the graph of the function uλsubscript𝑢𝜆u_{\lambda}italic_u start_POSTSUBSCRIPT italic_λ end_POSTSUBSCRIPT almost overlaps with the graph of uλexactsubscript𝑢subscript𝜆exactu_{\lambda_{\text{exact}}}italic_u start_POSTSUBSCRIPT italic_λ start_POSTSUBSCRIPT exact end_POSTSUBSCRIPT end_POSTSUBSCRIPT.

Refer to caption
Figure 2. Approximate solution uλsubscript𝑢𝜆u_{\lambda}italic_u start_POSTSUBSCRIPT italic_λ end_POSTSUBSCRIPT vs exact solution uλexactsubscript𝑢subscript𝜆exactu_{\lambda_{\text{exact}}}italic_u start_POSTSUBSCRIPT italic_λ start_POSTSUBSCRIPT exact end_POSTSUBSCRIPT end_POSTSUBSCRIPT
Remark 3.1.

The conclusion of Theorem 2.5 clearly remains valid under assumptions (h1)-(h3), if instead of the functionals ψλsubscript𝜓𝜆\psi_{\lambda}italic_ψ start_POSTSUBSCRIPT italic_λ end_POSTSUBSCRIPT we consider the functionals

ψ¯λ(u)=0θ(λ)|u(s)|𝑑s,subscript¯𝜓𝜆𝑢superscriptsubscript0𝜃𝜆𝑢𝑠differential-d𝑠\bar{\psi}_{\lambda}(u)=\int_{0}^{\theta(\lambda)}|u(s)|\,ds,over¯ start_ARG italic_ψ end_ARG start_POSTSUBSCRIPT italic_λ end_POSTSUBSCRIPT ( italic_u ) = ∫ start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_θ ( italic_λ ) end_POSTSUPERSCRIPT | italic_u ( italic_s ) | italic_d italic_s ,

and instead of φ𝜑\varphiitalic_φ we correspondingly take

φ~(λ)=0θ(λ)|uλ(s)|𝑑s.~𝜑𝜆superscriptsubscript0𝜃𝜆subscript𝑢𝜆𝑠differential-d𝑠\tilde{\varphi}(\lambda)=\int_{0}^{\theta(\lambda)}|u_{\lambda}(s)|\,ds.over~ start_ARG italic_φ end_ARG ( italic_λ ) = ∫ start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_θ ( italic_λ ) end_POSTSUPERSCRIPT | italic_u start_POSTSUBSCRIPT italic_λ end_POSTSUBSCRIPT ( italic_s ) | italic_d italic_s .

Moreover, we can extend this result to Lpsubscript𝐿𝑝L_{p}italic_L start_POSTSUBSCRIPT italic_p end_POSTSUBSCRIPT type functionals of the form

ψ~λ(u)=(0θ(λ)|u(s)|p𝑑s)1p,subscript~𝜓𝜆𝑢superscriptsuperscriptsubscript0𝜃𝜆superscript𝑢𝑠𝑝differential-d𝑠1𝑝\tilde{\psi}_{\lambda}(u)=\left(\int_{0}^{\theta(\lambda)}|u(s)|^{p}\,ds\right% )^{\frac{1}{p}},over~ start_ARG italic_ψ end_ARG start_POSTSUBSCRIPT italic_λ end_POSTSUBSCRIPT ( italic_u ) = ( ∫ start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_θ ( italic_λ ) end_POSTSUPERSCRIPT | italic_u ( italic_s ) | start_POSTSUPERSCRIPT italic_p end_POSTSUPERSCRIPT italic_d italic_s ) start_POSTSUPERSCRIPT divide start_ARG 1 end_ARG start_ARG italic_p end_ARG end_POSTSUPERSCRIPT ,

where 1p<1𝑝1\leq p<\infty1 ≤ italic_p < ∞. In this case, if we replace (h3) by condition

  1. (h3’)

    There exists a constant a>p𝑎𝑝a>pitalic_a > italic_p such that, for all λ>0𝜆0\lambda>0italic_λ > 0, one has

    |f(t,x,λ)||x|a(θ(λ)t)+Cλ,𝑓𝑡𝑥𝜆𝑥𝑎𝜃𝜆𝑡subscript𝐶𝜆\left|f(t,x,\lambda)\right|\leq\frac{|x|}{a(\theta(\lambda)-t)}+C_{\lambda},| italic_f ( italic_t , italic_x , italic_λ ) | ≤ divide start_ARG | italic_x | end_ARG start_ARG italic_a ( italic_θ ( italic_λ ) - italic_t ) end_ARG + italic_C start_POSTSUBSCRIPT italic_λ end_POSTSUBSCRIPT ,

    for all t[0,θ(λ))𝑡0𝜃𝜆t\in[0,\theta(\lambda))italic_t ∈ [ 0 , italic_θ ( italic_λ ) ) and all x𝑥x\in\mathbb{R}italic_x ∈ blackboard_R, where Cλ0subscript𝐶𝜆0C_{\lambda}\geq 0italic_C start_POSTSUBSCRIPT italic_λ end_POSTSUBSCRIPT ≥ 0 and the map λCλmaps-to𝜆subscript𝐶𝜆\lambda\mapsto C_{\lambda}italic_λ ↦ italic_C start_POSTSUBSCRIPT italic_λ end_POSTSUBSCRIPT is continuous.

4. Extension to fractional differential equations

The above results can be generalized to fractional differential equations with moving singularities. Such problems more accurately describe various physical, biological, or medical processes (see, e.g., [20, 21, 22, 23]). Therefore, our results related to problem (1.1) can be extended to the following problem:

{Dαcu(t)=f(t,u(t),λ),t[0,θ(λ))u(0)=u0(λ),casesformulae-sequencesuperscriptsuperscript𝐷𝛼𝑐𝑢𝑡𝑓𝑡𝑢𝑡𝜆𝑡0𝜃𝜆otherwise𝑢0subscript𝑢0𝜆otherwise\displaystyle\begin{cases}{}^{c}D^{\alpha}u(t)=f(t,u(t),\lambda),\quad t\in[0,% \theta(\lambda))\\ u(0)=u_{0}(\lambda),\end{cases}{ start_ROW start_CELL start_FLOATSUPERSCRIPT italic_c end_FLOATSUPERSCRIPT italic_D start_POSTSUPERSCRIPT italic_α end_POSTSUPERSCRIPT italic_u ( italic_t ) = italic_f ( italic_t , italic_u ( italic_t ) , italic_λ ) , italic_t ∈ [ 0 , italic_θ ( italic_λ ) ) end_CELL start_CELL end_CELL end_ROW start_ROW start_CELL italic_u ( 0 ) = italic_u start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT ( italic_λ ) , end_CELL start_CELL end_CELL end_ROW

where Dαcsuperscriptsuperscript𝐷𝛼𝑐{}^{c}D^{\alpha}start_FLOATSUPERSCRIPT italic_c end_FLOATSUPERSCRIPT italic_D start_POSTSUPERSCRIPT italic_α end_POSTSUPERSCRIPT is the Caputo fractional derivative and 0<α<10𝛼10<\alpha<10 < italic_α < 1. As shown in the literature, the above problem is equivalent to the Voltera integral equation

uλ(t)=u0(λ)+1Γ(α)0t(ts)α1f(s,u(s),λ)𝑑s.subscript𝑢𝜆𝑡subscript𝑢0𝜆1Γ𝛼superscriptsubscript0𝑡superscript𝑡𝑠𝛼1𝑓𝑠𝑢𝑠𝜆differential-d𝑠u_{\lambda}(t)=u_{0}(\lambda)+\frac{1}{\Gamma(\alpha)}\int_{0}^{t}(t-s)^{% \alpha-1}f(s,u(s),\lambda)ds.italic_u start_POSTSUBSCRIPT italic_λ end_POSTSUBSCRIPT ( italic_t ) = italic_u start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT ( italic_λ ) + divide start_ARG 1 end_ARG start_ARG roman_Γ ( italic_α ) end_ARG ∫ start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_t end_POSTSUPERSCRIPT ( italic_t - italic_s ) start_POSTSUPERSCRIPT italic_α - 1 end_POSTSUPERSCRIPT italic_f ( italic_s , italic_u ( italic_s ) , italic_λ ) italic_d italic_s .

Note that, in our case, the control problem remain unchanged, i.e., find λ𝜆\lambdaitalic_λ such that

0θ(λ)uλ(s)𝑑s=p.superscriptsubscript0𝜃𝜆subscript𝑢𝜆𝑠differential-d𝑠𝑝\int_{0}^{\theta(\lambda)}u_{\lambda}(s)ds=p.∫ start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_θ ( italic_λ ) end_POSTSUPERSCRIPT italic_u start_POSTSUBSCRIPT italic_λ end_POSTSUBSCRIPT ( italic_s ) italic_d italic_s = italic_p .

Since our entire analysis is grounded in the integral form of the Cauchy problem, we can easily extend the proof steps to address this more general case. The flexibility of the integral formulation allows for the adaptation of our methods without significant modifications. By imposing conditions similar to those outlined in (h1) and (h2), we can rigorously establish the controllability of the problem.

5. Conclusions

The analyzed control problem in this paper is atypical in several aspects: (a) it refers to equations with singularity; (b) the singularity itself depends on the control variable; (c) the controllability condition involves the moving singularity. All these aspects make the analysis much more complex and adapted to the specifics of the problem. The working techniques can also be taken into account for the investigation of other types of singular equations and controllability conditions including singular partial differential equations (see, e.g., [6]). We believe and anticipate that the ideas and techniques used in this article will have the high degree of suitability for the specifics of each individual problem in future research.

References

  • [1] Torres, P.J. Mathematical Models with Singularities: A Zoo of Singular Creatures; Atlantis Press, Amsterdam, 2015.
  • [2] O’Regan, D. Theory of Singular Boundary Value Problems; World Scientific Publishing, River Edge, NJ,. USA, 1994.
  • [3] Callegari, A; Nachman, A. A nonlinear singular boundary value problem in the theory of pseudoplasitc fluids. SIAM J. Appl. Math. 1980, 38, 275–282.
  • [4] Gingold, H. Rosenblat, S. Differential equations with moving singularities. SIAM J. Math. Anal 1976, 7(6), 942–957.
  • [5] Gingold, H. Introduction to differential equations with moving singularities. Rocky Mountain J. Math. 1976, 6(4), 571–574.
  • [6] Fila, M; Takahashi, J; Yanagida, E. Solutions with moving singularities for a one-dimensional nonlinear diffusion equation. Math. Ann. 2024, https://doi.org/10.1007/s00208-024-02882-0.
  • [7] Nicolescu, M; Dinculeanu, N; Marcus, S. Manual de analiză matematicăa (II); Editura Didactică şi Pedagogică: Bucureşti, Romania, 1964. (In Romanian)
  • [8] Haplea, I.Ş; Parajdi, L.G; Precup, R. On the controllability of a system modeling cell dynamics related to leukemia. Symmetry 2021, 13, 1867.
  • [9] Precup, R. On some applications of the controllability principle for fixed point equations. Results Appl. Math. 2022, 13, 100236.
  • [10] Coron, J.M. Control and Nonlinearity; AMS: Providence, 2007.
  • [11] Coddington, E.A. An Introduction to Ordinary Differential Equations; Dover: New York, 1961.
  • [12] Precup, R. Ordinary Differential Equations; De Gruyter: Berlin, 2018.
  • [13] O’Regan, D; Precup, R. Theorems of Leray-Schauder Type and Applications; CRC Press, 2002.
  • [14] Ciarlet, G. Linear and Nonlinear Functional Analysis with Applications; SIAM: Philadelphia, 2013.
  • [15] Le Dret, H. Nonlinear Elliptic Partial Differential Equations; Springer: Berlin, 2018.
  • [16] Gronwall, T.H. Note on the derivatives with respect to a parameter of the solutions of a system of differential equations. Ann. Math. 1919, 20(4), 292–296.
  • [17] Pachpatte, B.G.; Ames, W.F. Inequalities for Differential and Integral Equations; Academic Press, 1997.
  • [18] Rudin, W. Principles of Mathematical Analysis; MacGraw-Hill, 1976.
  • [19] Parajdi, L.G; Precup, R; Haplea, I.S. A method of lower and upper solutions for control problems and application to a model of bone marrow transplantation. Int. J. Appl. Math. Comput. Sci. 2023, 33(3), 409 - 418.
  • [20] Samko, S.G.; Kilbas, A.A.; Marichev, O.I. Fractional Integrals and Derivatives: Theory and Applications; Gordon and Breach Science Publishers: Yverdon, 1993.
  • [21] Kilbas, A.A.; Trujillo, J.J. Differentiale equations of fractional order: methods, results and problems-I. Appl. Anal. 2001, 78, 153–192.
  • [22] Benchohra, M.; Hamani, S.; Ntouyas, S.K. Boundary value problems for differential equations with fractional order and nonlocal conditions. Nonlinear Anal. 2009, 71(7-8), 2391–2396.
  • [23] Podlubny, I. Fractional Differential Equations; Mathematics in Sciences and Engineering, Vol. 198; Academic Press: San Diego, 1999.
2025

Related Posts