Conditions are derived of the existence of solutions of nonlinear boundary-value problems for systems of n ordinary differential equations with constant coefficients and single delay (in the linear part) and with a finite number of measurable delays of argument in nonlinearity: ż(t)=Az(t-τ)+g(t)+εZ(z(hi(t),t,ε), t∈[a,b], assuming that these solutions satisfy the initial and boundary conditions z(s):=ψ(s)ifs∉[a,b], ℓz(·)=α∈ℝm. The use of a delayed matrix exponential and a method of pseudoinverse by Moore-Penrose matrices led to an explicit and analytical form of sufficient conditions for the existence of solutions in a given space and, moreover, to the construction of an iterative process for finding the solutions of such problems in a general case when the number of boundary conditions (defined by a linear vector functional ℓ) does not coincide with the number of unknowns in the differential system with a single delay.
1. Introduction
First, we derive some auxiliary results concerning the theory of differential equations with delay. Consider a system of linear differential equations with concentrated delayż(t)-A(t)z(h0(t))=g(t)ift∈[a,b],
assuming thatz(s):=ψ(s)ifs∉[a,b],
where A is an n×n real matrix and g is an n-dimensional real column-vector with components in the space Lp[a,b] (where p∈[1,∞)) of functions summable on [a,b]; the delay h0(t)≤t is a function h0:[a,b]→ℝ measurable on [a,b]; ψ:ℝ∖[a,b]→ℝn is a given function. Using the denotations(Sh0z)(t):={z(h0(t))ifh0(t)∈[a,b],θifh0(t)∉[a,b],ψh0(t):={θifh0(t)∈[a,b],ψ(h0(t))ifh0(t)∉[a,b],
where θ is an n-dimensional zero column-vector and assuming t∈[a,b], it is possible to rewrite (1.1), (1.2) as(Lz)(t):=ż(t)-A(t)(Sh0z)(t)=φ(t),t∈[a,b],
where φ is an n-dimensional column-vector defined by the formulaφ(t):=g(t)+A(t)ψh0(t)∈Lp[a,b].
We will investigate (1.5) assuming that the operator L maps a Banach space Dp[a,b] of absolutely continuous functions z:[a,b]→ℝn into a Banach space Lp[a,b](1≤p<∞) of functions φ:[a,b]→ℝn summable on [a,b]; the operator Sh0 maps the space Dp[a,b] into the space Lp[a,b]. Transformations (1.3), (1.4) make it possible to add the initial function ψ(s), s<a to nonhomogeneity generating an additive and homogeneous operation not depending on ψ and without the classical assumption regarding the continuous connection of solution z(t) with the initial function ψ(t) at the point t=a.
A solution of differential system (1.5) is defined as an n-dimensional column vector-function z∈Dp[a,b], absolutely continuous on [a,b], with a derivative ż∈Lp[a,b] satisfying (1.5) almost everywhere on [a,b].
Such approach makes it possible to apply well-developed methods of linear functional analysis to (1.5) with a linear and bounded operator L. It is well-known (see: [1, 2]) that a nonhomogeneous operator equation (1.5) with delayed argument is solvable in the space Dp[a,b] for an arbitrary right-hand side φ∈Lp[a,b] and has an n-dimensional family of solutions (dimkerL=n) in the formz(t)=X(t)c+∫abK(t,s)φ(s)ds∀c∈Rn,
where the kernel K(t,s) is an n×n Cauchy matrix defined in the square [a,b]×[a,b] being, for every fixed s≤t, a solution of the matrix Cauchy problem (LK(⋅,s))(t):=∂K(t,s)∂t-A(t)(Sh0K(⋅,s))(t)=Θ,K(s,s)=I,
where K(t,s)≡Θ if a≤t<s≤b, Θ is n×n null matrix and I is n×n identity matrix. A fundamental n×n matrix X(t) for the homogeneous (φ≡θ) equation (1.5) has the form X(t)=K(t,a), X(a)=I [2]. Throughout the paper, we denote by Θs an s×s null matrix if s≠n, by Θs,p an s×p null matrix, by Is an s×s identity matrix if s≠n, and by θs an s-dimensional zero column-vector if s≠n.
A serious disadvantage of this approach, when investigating the above-formulated problem, is the necessity to find the Cauchy matrix K(t,s) [3, 4]. It exists but, as a rule, can only be found numerically. Therefore, it is important to find systems of differential equations with delay such that this problem can be solved directly. Below we consider the case of a system with so-called single delay [5]. In this case, the problem of how to construct the Cauchy matrix is successfully solved analytically due to a delayed matrix exponential defined below.
1.1. A Delayed Matrix Exponential
Consider a Cauchy problem for a linear nonhomogeneous differential system with constant coefficients and with a single delay τż(t)=Az(t-τ)+g(t),z(s)=ψ(s),ifs∈[-τ,0],
with an n×n constant matrix A, g:[0,∞)→ℝn, ψ:[-τ,0]→ℝn, τ>0 and an unknown vector-solution z:[-τ,∞)→ℝn. Together with a nonhomogeneous problem (1.9), (1.10), we consider a related homogeneous problem ż(t)=Az(t-τ),z(s)=ψ(s),ifs∈[-τ,0].
Denote by eτAt a matrix function called a delayed matrix exponential (see [5]) and defined aseτAt:={Θif-∞<t<-τ,Iif-τ≤t<0,I+At1!if0≤t<τ,I+At1!+A2(t-τ)22!ifτ≤t<2τ,⋯I+At1!+⋯+Ak(t-(k-1)τ)kk!if(k-1)τ≤t<kτ,⋯.
This definition can be reduced to the following expression: eτAt=∑n=0[t/τ]+1An(t-(n-1)τ)nn!,
where [t/τ] is the greatest integer function. The delayed matrix exponential equals the unit matrix I on [-τ,0] and represents a fundamental matrix of a homogeneous system with single delay. Thus, the delayed matrix exponential solves the Cauchy problem for a homogeneous system (1.11), satisfying the unit initial conditions z(s)=ψ(s)≡eτAs=Iif-τ≤s≤0,
and the following statement holds (see, e.g., [5], [6, Remark 1], [7, Theorem 2.1]).
Lemma 1.1.
A solution of a Cauchy problem for a nonhomogeneous system with single delay (1.9), satisfying a constant initial condition
z(s)=ψ(s)=c∈Rnifs∈[-τ,0]
has the form
z(t)=eτA(t-τ)c+∫0teτA(t-τ-s)g(s)ds.
The delayed matrix exponential was applied, for example, in [6, 7] to investigation of boundary value problems of diffferential systems with a single delay and in [8] to investigation of the stability of linear perturbed systems with a single delay.
1.2. Fredholm Boundary-Value Problem
Without loss of generality, let a=0 and, with a view of the above, the problem (1.9), (1.10) can be transformed (h0(t):=t-τ) to an equation of the type (1.1) (see (1.5))ż(t)-A(Sh0z)(t)=φ(t),t∈[0,b],
where, in accordance with (1.3),(1.4)(Sh0z)(t)={z(t-τ)ift-τ∈[0,b],θift-τ∉[0,b],φ(t)=g(t)+Aψh0(t)∈Lp[0,b],ψh0(t)={θift-τ∈[0,b],ψ(t-τ)ift-τ∉[0,b].
A general solution of problem (1.17) for a nonhomogeneous system with single delay and zero initial data has the form (1.7) z(t)=X(t)c+∫0bK(t,s)φ(s)ds∀c∈Rn,
where, as can easily be verified (in view of the above-defined delayed matrix exponential) by substituting into (1.17), X(t)=eτA(t-τ),X(0)=eτ-Aτ=I
is a normal fundamental matrix of the homogeneous system related to (1.17) (or (1.9)) with initial data X(0)=I, and the Cauchy matrix K(t,s) has the form K(t,s)=eτA(t-τ-s)if0≤s<t≤b,K(t,s)≡Θif0≤t<s≤b.
Obviously K(t,0)=eτA(t-τ)=X(t),K(0,0)=eτA(-τ)=X(0)=I,
and, therefore, the initial problem (1.17) for systems of ordinary differential equations with constant coefficients and single delay has an n-parametric family of linearly independent solutions (1.16).
Now, we will deal with a general boundary-value problem for system (1.17). Using the results [2, 9], it is easy to derive statements for a general boundary-value problem if the number m of boundary conditions does not coincide with the number n of unknowns in a differential system with single delay.
We consider a boundary-value problem ż(t)-Az(t-τ)=g(t),t∈[0,b],z(s):=ψ(s),s∉[0,b],
assuming that lz(⋅)=α∈Rm,
or, using (1.18), its equivalent formż(t)-A(Sh0z)(t)=φ(t),t∈[0,b],lz(⋅)=α∈Rm,
where α is an m-dimensional constant vector-column ℓ is an m-dimensional linear vector-functional defined on the space Dp[0,b] of an n-dimensional vector-functions l=col(l1,…,lm):Dp[0,b]⟶Rm,li:Dp[0,b]⟶R,i=1,…,m,
absolutely continuous on [0,b]. Such problems for functional-differential equations are of Fredholm's type (see, e.g., [1, 2]). In order to formulate the following result, we need several auxiliary abbreviations. We set Q:=lX(⋅)=leτA(⋅-τ).
We define an n×n-dimensional matrix (orthogonal projection) PQ:=I-Q+Q,
projecting space ℝn to kerQ of the matrix Q.
Moreover, we define an m×m-dimensional matrix (orthogonal projection) PQ*:=Im-QQ+,
projecting space ℝm to kerQ* of the transposed matrix Q*=QT, where Im is an m×m identity matrix and Q+ is an n×m-dimensional matrix pseudoinverse to the m×n-dimensional matrix Q. Denote d:=rankPQ* and n1:=rankQ=rankQ*. Since rankPQ*=m-rankQ*,
we have d=m-n1.
We will denote by PQd* an d×m-dimensional matrix constructed from d linearly independent rows of the matrix PQ*. Denote r:=rankPQ. Since rankPQ=n-rankQ,
we have r=n-n1. By PQr we will denote an n×r-dimensional matrix constructed from r linearly independent columns of the matrix PQ. Finally, we define Xr(t):=X(t)PQr,
and a generalized Green operator (Gφ)(t):=∫0bG(t,s)φ(s)ds,
where G(t,s):=K(t,s)-eτA(t-τ)Q+lK(⋅,s)
is a generalized Green matrix corresponding to the boundary-value problem (1.25) (the Cauchy matrix K(t,s) has the form (1.21)).
In [6, Theorem 4], the following result (formulating the necessary and sufficient conditions of solvability and giving representations of the solutions z∈Dp[0,b], ż∈Lp[0,b] of the boundary-value problem (1.25) in an explicit analytical form) is proved.
Theorem 1.2.
If n1≤min(m,n), then:
the homogeneous problem
ż(t)-A(Sh0z)(t)=θ,t∈[0,b],lz(⋅)=θm∈Rm
corresponding to problem (1.25) has exactly r linearly independent solutions
z(t,cr)=Xr(t)cr=eτA(t-τ)PQrcr∈Dp[0,b],
nonhomogeneous problem (1.25) is solvable in the space Dp[0,b] if and only if φ∈Lp[0,b] and α∈ℝm satisfy d linearly independent conditions
PQd*⋅(α-l∫0bK(⋅,s)φ(s)ds)=θd,
in that case the nonhomogeneous problem (1.25) has an r-dimensional family of linearly independent solutions represented in an analytical form
z(t)=z0(t,cr):=Xr(t)cr+(Gφ)(t)+X(t)Q+α∀cr∈Rr.
2. Perturbed Weakly Nonlinear Boundary Value Problems
As an example of applying Theorem 1.2, we consider a problem of the branching of solutions z:[0,b]→ℝn, b>0 of systems of nonlinear ordinary differential equations with a small parameter ɛ and with a finite number of measurable delays hi(t), i=1,2,…,k of argument of the formż(t)=Az(t-τ)+g(t)+ɛZ(z(hi(t)),t,ɛ),t∈[0,b],hi(t)≤t,
satisfying the initial and boundary conditionsz(s)=ψ(s),ifs<0,lz(⋅)=α,α∈Rm,
and such that its solution z=z(t,ɛ), satisfying z(⋅,ɛ)∈Dp[0,b],ż(⋅,ɛ)∈Lp[0,b],z(t,⋅)∈C[0,ɛ0],
for a sufficiently small ɛ0>0, for ɛ=0, turns into one of the generating solutions (1.38); that is, z(t,0)=z0(t,cr) for a cr∈ℝr. We assume that the n×1 vector-operator Z satisfies Z(⋅,t,ɛ)∈C1[‖z-z0‖≤q],Z(z(hi(t)),⋅,ɛ)∈Lp[0,b],Z(z(hi(t)),t,⋅)∈C[0,ɛ0],
where q>0 is sufficiently small. Using denotations (1.3), (1.4), and (1.6), it is easy to show that the perturbed nonlinear boundary value problem (2.1), (2.2) can be rewritten in the formż(t)=A(Sh0z)(t)+ɛZ((Shz)(t),t,ɛ)+φ(t),lz(⋅)=α,t∈[0,b].
In (2.5), A is an n×n constant matrix, h0:[0,b]→ℝ is a single delay defined by h0(t):=t-τ, τ>0, (Shz)(t)=col[(Sh1z)(t),…,(Shkz)(t)]
is an N-dimensional column vector, where N=nk, and φ is an n-dimensional column vector given by φ(t)=g(t)+Aψh0(t).
The operator Sh maps the space Dp into the space LpN=Lp×⋯×Lp︸k-times,
that is, Sh:Dp→LpN. Using denotation (1.3) for the operator Shi:Dp→Lp, i=1,…,k, we have the following representation:(Shiz)(t)=∫0bχhi(t,s)ż(s)ds+χhi(t,0)z(0),
where χhi(t,s)={1,if(t,s)∈Ωi,0,if(t,s)∉Ωi
is the characteristic function of the set Ωi:={(t,s)∈[0,b]×[0,b]:0≤s≤hi(t)≤b}.
Assume that the generating boundary value problemż(t)=A(Sh0z)(t)+φ(t),lz=α,
being a particular case of (2.5) for ɛ=0, has solutions for nonhomogeneities φ∈Lp[0,b] and α∈ℝm that satisfy conditions (1.37). In such a case, by Theorem 1.2, the problem (2.12) possesses an r-dimensional family of solutions of the form (1.38).
Problem 1.
Below, we consider the following problem: derive the necessary and sufficient conditions indicating when solutions of (2.5) turn into solutions (1.38) of the boundary value problem (2.12) for ɛ=0.
Using the theory of generalized inverse operators [2], it is possible to find conditions for the solutions of the boundary value problem (2.5) to be branching from the solutions of (2.5) with ɛ=0. Below, we formulate statements, solving the above problem. As compared with an earlier result [10, page 150], the present result is derived in an explicit analytical form. The progress was possible by using the delayed matrix exponential since, in such a case, all the necessary calculations can be performed to the full.
Theorem 2.1 (necessary condition).
Consider the system (2.1); that is,
ż(t)=Az(t-τ)+g(t)+εZ(z(hi(t)),t,ε),t∈[0,b],
where hi(t)≤t, i=1,…,k, with the initial and boundary conditions (2.2); that is,
z(s)=ψ(s),ifs<0<b,lz(⋅)=α∈Rm,
and assume that, for nonhomogeneities
φ(t)=g(t)+Aψh0(t)∈Lp[0,b],
and for α∈ℝm, the generating boundary value problem
ż(t)=A(Sh0z)(t)+φ(t),lz(⋅)=α,
corresponding to the problem (1.25), has exactly an r-dimensional family of linearly independent solutions of the form (1.38). Moreover, assume that the boundary value problem (2.13), (2.14) has a solution z(t,ɛ) which, for ɛ=0, turns into one of solutions z0(t,cr) in (1.38) with a vector-constant cr:=cr0∈ℝr.
Then, the vector cr0 satisfies the equationF(cr0):=∫0bH(s)Z((Shz0)(s,cr0),s,0)ds=θd,
where
H(s):=PQd*lK(⋅,s)=PQd*leτA(⋅-τ-s).
Proof.
We consider the nonlinearity in system (2.13), that is, the term ɛZ(z(hi(t)),t,ɛ) as an inhomogeneity, and use Theorem 1.2 assuming that condition (1.37) is satisfied. This gives
∫0bH(s)Z((Shz)(s,ɛ),s,ɛ)ds=θd.
In this integral, letting ɛ→0, we arrive at the required condition (2.17).
Corollary 2.2.
For periodic boundary-value problems, the vector-constant cr∈ℝr has a physical meaning-it is the amplitude of the oscillations generated. For this reason, (2.17) is called an equation generating the amplitude [11]. By analogy with the investigation of periodic problems, it is natural to say (2.17) is an equation for generating the constants of the boundary value problem (2.13), (2.14). If (2.17) is solvable, then the vector constant cr0∈ℝr specifies the generating solution z0(t,cr0) corresponding to the solution z=z(t,ɛ) of the original problem such that
z(⋅,ɛ):[0,b]⟶Rn,z(⋅,ɛ)∈Dp[0,b],ż(⋅,ɛ)∈Lp[0,b],z(t,⋅)∈C[0,ɛ0],z(t,0)=z0(t,cr0).
Also, if (2.17) is unsolvable, the problem (2.13), (2.14) has no solution in the analyzed space. Note that, here and in what follows, all expressions are obtained in the real form and hence, we are interested in real solutions of (2.17), which can be algebraic or transcendental.
Sufficient conditions for the existence of solutions of the boundary-value problem (2.13), (2.14) can be derived using results in [10, page 155] and [2]. By changing the variables in system (2.13), (2.14) z(t,ɛ)=z0(t,cr0)+y(t,ɛ),
we arrive at a problem of finding sufficient conditions for the existence of solutions of the problemẏ(t)=A(Sh0y)(t)+ɛZ(Sh(z0+y)(t),t,ɛ),ly=θm,t∈[0,b],
and such that y(⋅,ɛ):[0,b]⟶Rn,y(⋅,ɛ)∈Dp[0,b],ẏ(⋅,ɛ)∈Lp[0,b],y(t,⋅)∈C[0,ɛ0],y(t,0)=θ.
Since the vector function Z((Shz)(t),t,ɛ) is continuously differentiable with respect to z and continuous in ɛ in the neighborhood of the point (z,ɛ)=(z0(t,cr0),0),
we can separate its linear term as a function depending on y and terms of order zero with respect to ɛZ(Sh(z0(t,cr0)+y),t,ɛ)=f0(t,cr0)+A1(t)(Shy)(t)+R((Shy)(t),t,ɛ),
where f0(t,cr0):=Z((Shz0)(t,cr0),t,0),f0(⋅,cr0)∈Lp[0,b],A1(t)=A1(t,cr0)=∂Z(Shx,t,0)∂Shx|x=z0(t,cr0),A1(⋅)∈Lp[0,b],R(θ,t,0)=θ,∂R(θ,t,0)∂y=Θ,R(y,⋅,ɛ)∈Lp[0,b].
We now consider the vector function Z((Sh(z0+y))(t),t,ɛ) in (2.22) as an inhomogeneity and we apply Theorem 1.2 to this system. As the result, we obtain the following representation for the solution of (2.22): y(t,ɛ)=Xr(t)c+y(1)(t,ɛ).
In this expression, the unknown vector of constants c=c(ɛ)∈C[0,ɛ0] is determined from a condition similar to condition (1.37) for the existence of solution of problem (2.22):B0c=∫0bH(s)[A1(s)(Shy(1))(s,ɛ)+R((Shy)(s,ɛ),s,ɛ)]ds,
where B0=∫0bH(s)A1(s)(ShXr)(s)ds
is a d×r matrix, and H(s):=PQd*lK(⋅,s)=PQd*leτA(⋅-τ-s).
The unknown vector function y(1)(t,ɛ) is determined by using the generalized Green operator as follows: y(1)(t,ɛ)=ɛ(G[Z(Sh(z0(s,cr0)+y),s,ɛ)])(t).
Let PN(B0) be an r×r matrix orthoprojector ℝr→N(B0), and let PN(B0*) be a d×d matrix-orthoprojector ℝd→N(B0*). Equation (2.28) is solvable with respect to c∈ℝr if and only if PN(B0*)∫0bH(s)[A1(s)(Shy(1))(s,ɛ)+R((Shy)(s,ɛ),s,ɛ)]ds=θd.
For PN(B0*)=Θd,
the last condition is always satisfied and (2.28) is solvable with respect to c∈ℝr up to an arbitrary vector constant PN(B0)c∈ℝr from the null space of the matrix B0c=B0+∫0bH(s)[A1(s)(Shy(1))(s,ɛ)+R((Shy)(s,ɛ),s,ɛ)]ds+PN(B0)c.
To find a solution y=y(t,ɛ) of (2.28) such that y(⋅,ɛ):[0,b]⟶Rn,y(⋅,ɛ)∈Dp[0,b],ẏ(⋅,ɛ)∈Lp[0,b],y(t,⋅)∈C[0,ɛ0],y(t,0)=θ,
it is necessary to solve the following operator system:y(t,ɛ)=Xr(t)c+y(1)(t,ɛ),c=B0+∫0bH(s)[A1(s)(Shy(1))(s,ɛ)+R((Shy)(s,ɛ),s,ɛ)]ds,y(1)(t,ɛ)=ɛG[Z(Sh(z0(s,cr0)+y),s,ɛ)](t).
The operator system (2.36) belongs to the class of systems solvable by the method of simple iterations, convergent for sufficiently small ɛ∈[0,ɛ0] (see [10, page 188]). Indeed, system (2.36) can be rewritten in the formu=L(1)u+Fu,
where u=col(y(t,ɛ),c(ɛ),y(1)(t,ɛ)) is a (2n+r)-dimensional column vector, L(1) is a linear operator L(1):=(ΘXr(t)IΘr,nΘr,rL1ΘΘn,rΘ),
where L1(*)=B0+∫0bH(s)A1(s)(*)ds,
and F is a nonlinear operator Fu:=(θB0+∫0bH(s)R((Shy)(s,ɛ),s,ɛ)dsɛ(G[Z((Shz0)(s,cr0),s,0)+A1(s)(Shy)(s,ɛ)+R((Shy)(s,ɛ),s,ɛ)])(t)).
In view of the structure of the operator L(1) containing zero blocks on and below the main diagonal, the inverse operator (I2n+r-L(1))-1
exists. System (2.37) can be transformed into u=Su,
where S:=(I2n+r-L(1))-1F
is a contraction operator in a sufficiently small neighborhood of the point (z,ɛ)=(z0(t,cr0),0).
Thus, the solvability of the last operator system can be established by using one of the existing versions of the fixed-point principles [12] applicable to the system for sufficiently small ɛ∈[0,ɛ0]. It is easy to prove that the sufficient condition PN(B0*)=Θd for the existence of solutions of the boundary value problem (2.13), (2.14) means that the constant cr0∈ℝr of the equation for generating constant (2.17) is a simple root of equation (2.17) [2]. By using the method of simple iterations, we can find the solution of the operator system and hence the solution of the original boundary value problem (2.13), (2.14). Now, we arrive at the following theorem.
Theorem 2.3 (sufficient condition).
Assume that the boundary value problem (2.13), (2.14) satisfies the conditions listed above and the corresponding linear boundary value problem (1.25) has an r-dimensional family of linearly independent solutions of the form (1.38). Then, for any simple root cr=cr0∈ℝr of the equation for generating the constants (2.17), there exist at least one solution of the boundary value problem (2.13), (2.14). The indicated solution z(t,ɛ) is such that
z(⋅,ɛ)∈Dp[0,b],ż(⋅,ɛ)∈Lp[0,b],z(t,⋅)∈C[0,ɛ0],
and, for ɛ=0, turns into one of the generating solutions (1.38) with a constant cr0∈ℝr; that is, z(t,0)=z0(t,cr0). This solution can be found by the method of simple iterations, which is convergent for a sufficiently small ɛ∈[0,ɛ0].
Corollary 2.4.
If the number n of unknown variables is equal to the number m of boundary conditions (and hence r=d), the boundary value problem (2.13), (2.14) has a unique solution. In such a case, the problems considered for functional-differential equations are of Fredholm's type with a zero index. By using the procedure proposed in [2] with some simplifying assumptions, we can generalize the proposed method to the case of multiple roots of equation (2.17) to determine sufficient conditions for the existence of solutions of the boundary-value problem (2.13), (2.14).
3. Example
We will illustrate the above proved theorems on the example of a weakly perturbed linear boundary value problem. Consider the following simplest boundary value problem-a periodic problem for the delayed differential equation:ż(t)=z(t-τ)+ɛ∑i=1kBi(t)z(hi(t))+g(t),t∈(0,T],z(s)=ψ(s),ifs<0,z(0)=z(T),
where 0<τ,T=const, Bi are n×n matrices, Bi,g∈Lp[0,T], ψ:ℝ1∖(0,T]→ℝn, hi(t)≤t are measurable functions. Using the symbols Shi and ψhi (see (1.3), (1.4), (2.9)), we arrive at the following operator system: ż(t)=z(t-τ)+ɛB(t)(Shz)(t)+φ(t,ɛ),lz:=z(0)-z(T)=θn,
where B(t):=(B1(t),…,Bk(t)) is an n×N matrix (N=nk), and φ(t,ɛ):=g(t)+ψh0(t)+ɛ∑i=1kBi(t)ψhi(t)∈Lp[0,T].
We will consider the simplest case with T≤τ. Utilizing the delayed matrix exponential, it can be easily verified that in this case, the matrix X(t)=eτI(t-τ)=I
is a normal fundamental matrix for the homogeneous generating system ż(t)=z(t-τ).
Then, Q:=lX(⋅)=eτ-Iτ-eτI(T-τ)=θn,PQ=PQ*=I,(r=n,d=m=n),K(t,s)={eτI(t-τ-s)=I,0≤s≤t≤T,Θ,s>t,lK(⋅,s)=K(0,s)-K(T,s)=-I,H(τ)=PQ*lK(⋅,s)=-I,(ShiI)(t)=χhi(t,0)⋅I=I⋅{1,if0≤hi(t)≤T,0,ifhi(t)<0.
To illustrate the theorems proved above, we will find the conditions for which the boundary value problem (3.1) has a solution z(t,ɛ) that, for ɛ=0, turns into one of solutions (1.38) z0(t,cr) of the generating problem. In contrast to the previous works [7, 9], we consider the case when the unperturbed boundary-value problem ż(t)=z(t-τ)+φ(t,0),z(0)=z(T)
has an n-parametric family of linear-independent solutions of the form(1.38)z:=z0(t,cn)=cn+(Gφ)(t),∀cn∈Rn.
For this purpose, it is necessary and sufficient for the vector function φ(t)=g(t)+ψh0(t)
to satisfy the condition of type (1.37)∫0TH(s)φ(s)ds=-∫0Tφ(s)ds=θn.
Then, according to the Theorem 2.1, the constant cn=cn0∈ℝn must satisfy (2.17), that is, the equationF(cn0):=∫0TH(s)Z((Shz0)(s,cn0),s,0)ds=θn,
which in our case is a linear algebraic system B0cn0=-∫0TB(s)(Sh(Gφ))(s)ds,
with the n×n matrix B0 in the form B0=∫0TH(s)B(s)(ShI)(s)ds=-∫0T∑i=1kBi(s)(ShiI)(s)ds=-∑i=1k∫0TBi(s)χhi(s,0)ds.
According to Corollary 2.4, if detB0≠0, the problem (3.1) for the case T≤τ has a unique solution z(t,ɛ) with the properties z(⋅,ɛ)∈Dpn[0,T],ż(⋅,ɛ)∈Lpn[0,T],z(t,⋅)∈C[0,ɛ0],z(t,0)=z0(t,cn0),
for g∈Lp[0,T], ψ(t)∈Lp[0,T], and for measurable delays hi that which satisfy the criterion (3.10) of the existence of a generating solution where cn0=-B0+∫0TB(s)(Sh(Gφ))(s)ds.
A solution z(t,ɛ) of the boundary value problem (3.1) can be found by the convergent method of simple iterations (see Theorem 2.3).
If, for example, hi(t)=t-Δi, where 0<Δi=const<T, i=1,…,k, then χhi(t,0)={1if0≤hi(t)=t-Δi≤T,0ifhi(t)=t-Δi<0,={1ifΔi≤t≤T+Δi,0,ift<Δi.
The n×n matrix B0 can be rewritten in the form B0=∫0TH(s)∑i=1kBi(s)χhi(s,0)dτ=-∑i=1k∫0TBi(s)χhi(s,0)ds=-∑i=1k∫ΔiTBi(s)ds,
and the unique solvability condition of the boundary value problem (3.1) takes the form det[∑i=1k∫ΔiTBi(s)ds]≠0.
It is easy to see that if the vector function Z(z(hi(t)),t,ɛ) is nonlinear in z, for example as a square, then (3.11) generating the constants will be a square-algebraic system and, in this case, the boundary value problem (3.1) can have two solutions branching from the point ɛ=0.
Acknowledgments
The first and the fourth authors were supported by the Grant no. 1/0090/09 of the Grant Agency of Slovak Republic (VEGA) and by the project APVV-0700-07 of Slovak Research and Development Agency. The second author was supported by the Grant no. P201/11/0768 of Czech Grant Agency, by the Council of Czech Government MSM 0021630503 and by the Project FEKT/FSI-S-11-1-1159. The third author was supported by the Project no. M/34-2008 of Ukrainian Ministry of Education, Ukraine.
AzbelevN. V.MaksimovV. P.Equations with delayed arguments19831814191441translated in Differentsial'nye Uravneniya, Vol. 18, no. 12, pp. 2027-2050683997BoichukA. A.SamoilenkoA. M.2004Boston, Mass, USAVSP3172124880HaleJ.19772ndNew York, NY, USASpringer365Applied Mathematical Sciences, Vol. 30508721Mallet-ParetJ.The fredholm alternative for functional-differential equations of mixed type1999111147168046310.1023/A:1021889401235ZBL0927.34049KhusainovD. Ya.ShuklinG. V.Relative controllability in systems with pure delay2005412210221translated in Prikladnaya Mekhanika, Vol. 41, no. 2, pp. 118-13010.1007/s10778-005-0079-32190935BoichukA. A.DiblíkJ.KhusainovD.RůžičkováM.Fredholm's boundary-value problems for differential systems with a single delay20107252251225810.1016/j.na.2009.10.025ZBL1190.34073BoichukA. A.DiblíkJ.KhusainovD. Y. A.RůžičkováM.Boundary value problems for delay differential systems20102010105938342660943ZBL1204.34087MedvĕdM.PospíšilM.ŠkripkováL.Stability and the nonexistence of blowing-up solutions of nonlinear delay systems with linear parts defined by permutable matrices201174123903391110.1016/j.na.2011.02.026BoichukA. A.GrammatikopoulosM. K.Perturbed fredholm boundary value problems for delay differential systems20031584386410.1155/S10853375033040262010940ZBL1077.34069BoichukA. A.ZhuravlevV. F.SamoilenkoA. M.Generalized inverse operators and noether boundary-value problems13Proceedings of the Institute of Mathematics of the National Academy of Sciences of the Ukraine1995Kyiv, Ukrania3201367732MalkinI. G.1956Moscow, RussiaGostekhizdatKrasnosel'skiiM. A.VainikkoG. M.ZabreikoP. P.RutitskiiY. A. B.StetsenkoV. Ya.1968Moscow, RussiaNaukaTranslation: Noordhoff, Groningen, 19720385655