Oscillation results for first order neutral delay differential equations with several positive and negative coefficients Ábel Garab, Gergő Tóth Abstract We provide sufficient criteria for the oscillation of all solutions of neutral delay differential equations of the form [ x ⁡ ( t ) − ∑ i = 1 N r R i ​ ( t ) ​ x ​ ( t − r i ​ ( t ) ) ] ′ + ∑ i = 1 N p P i ​ ( t ) ​ x ​ ( t − τ i ​ ( t ) ) − ∑ i = 1 N q Q i ​ ( t ) ​ x ​ ( t − δ i ​ ( t ) ) = 0 , \left[x(t)-\sum_{i=1}^{N_{r}}R_{i}(t)x(t-r_{i}(t))\right]^{\prime}+\sum_{i=1}^{N_{p}}P_{i}(t)x(t-\tau_{i}(t))-\sum_{i=1}^{N_{q}}Q_{i}(t)x(t-\delta_{i}(t))=0, with both positive and negative terms and time-variable delays. Our results improve and generalize several existing criteria available in the literature that address restricted cases, such as constant delays or the absence of negative coefficients. Under additional assumptions on slowly varying parameters, we derive sharper oscillation conditions. We demonstrate the applicability of our findings through illustrative examples. 1 Introduction In this work, we consider linear neutral delay differential equations (NDDEs) with several variable delays of the form \left[x(t)-\sum_{i=1}^{N_{r}}R_{i}(t)x(t-r_{i}(t))\right]^{\prime}+\sum_{i=1}^{N_{p}}P_{i}(t)x(t-\tau_{i}(t))-\sum_{i=1}^{N_{q}}Q_{i}(t)x(t-\delta_{i}(t))=0, (1.1) for t\geq t_{0} , where R_{i},\ P_{j},\ Q_{k},\ r_{i},\ \tau_{j} , and \delta_{k} are continuous nonnegative functions, R_{i},\ \tau_{j},\ \delta_{k}\in C^{1} and r_{i},\ \tau_{j},\ \delta_{k}\leq D\in(0,\infty) on the interval [t_{0},\infty) for all 1\leq i\leq N_{r},\ 1\leq j\leq N_{p},\ 1\leq k\leq N_{q} , and N_{q}\leq N_{p} . Furthermore, we assume that \delta_{i}(t)\leq\tau_{i}(t) and \delta^{\prime}_{i}(t)<1 for all 1\leq i\leq N_{q} and t\in(t_{0},\infty) . Delay differential equations (DDEs) model systems where the current rate of change depends not only on the present state but also on past states. Among these, NDDEs form a subset in which the delayed term also appears in the derivative. NDDEs arise in many applications in biology [ 15 , 5 , 18 , 19 , 22 ] , in physics and engineering [ 20 , 40 , 34 , 33 , 35 ] , and even in economics [ 12 , 26 , 4 ] , to mention only a few examples. For a thorough introduction to the theory of DDEs, we refer to the books [ 17 ] and [ 7 ] . For a more applied approach, we recommend that the reader study [ 10 ] and [ 36 ] . The oscillation theory of DDEs has received considerable attention in the last sixty to seventy years. The monographs [ 16 , 9 , 1 ] yield a nice overview of the main contributions to the field from the last century. Each of these books devotes a complete chapter to oscillation and nonoscillation criteria for NDDEs. We say that a function is a solution of ( 1.1 ) if it is continuous on [t_{0}-D,\infty) , continuously differentiable on the interval (t_{0},\infty) , and satisfies ( 1.1 ) there. A solution of a scalar differential equation oscillates if it has arbitrarily large zeros. A differential equation oscillates if every solution of the equation oscillates. A solution x of ( 1.1 ) is eventually positive if there exists a T>t_{0} such that x(t)>0 for all t>T . A linear equation is oscillatory if and only if it has no eventually positive solution. Several studies have addressed oscillation criteria for special cases of equation ( 1.1 ). For instance, [ 16 , Section 6.9] considers time-variable delays and coefficient functions, but no negative terms are allowed (i.e. Q_{k}\equiv 0 for all k ), while [ 1 , Sections 3.3–3.5] involves both a single positive and a negative coefficient function (more precisely, N_{r}=N_{p}=N_{q}=1 ), and constant delays. There is still considerable progress related to the oscillation properties of equation ( 1.1 ). For instance, [ 2 ] studies the distance between consecutive zeros of solutions of similar equations; there are recent papers on the oscillation of dynamic equations [ 14 ] and nonlinear equations of a similar form [ 29 ] . However, very little is known about the oscillation of ( 1.1 ) in its full generality. The aim of this work is to fill this gap. We mention here the results of [ 28 , 25 , 11 ] that allow both positive and negative coefficient functions, but only constant delays. Their approach and the obtained oscillation criteria are not comparable to ours. Our first main result ( Theorem 3.1 ) generalizes the results of Győri and Ladas [ 16 , Section 6.9] . It establishes that the NDDE ( 1.1 ) is oscillatory, provided a corresponding first-order, non-neutral DDE is oscillatory. We combine this with classical oscillation criteria for the latter equation to obtain explicit criteria for the oscillation of ( 1.1 ). Our second main result ( Theorem 4.2 ) is based on a comparison-type result ( Theorem 4.1 ) and an iterative technique. It yields a generalization of a result by Agarwal, Bohner, and Li [ 1 , Theorem 3.4.9] for multiple constant delays. We demonstrate that both main theorems can be sharpened when a property related to slow variation is fulfilled (see Corollaries 3.4 and 4.3 ). The rest of the paper is structured as follows. In the next section, we clarify some further notations and prove two auxiliary results that are used throughout the paper. Section 3 contains our main results related to the general case, when the delays may be time-variable. Section 4 is devoted to our results related to the case of constant delays. We conclude the paper with some illustrative examples in Section 5 , where we also discuss the relation of the obtained oscillation criteria. 2 Preliminaries In this section, we set some notations and prove two auxiliary results. As customary, let \mathbb{N} and \mathbb{R} denote the set of nonnegative integers and reals, respectively. The next simple proposition is needed for Lemma 2.2 , which plays a key role in the proof of all our results. Proposition 2.1 . Assume that \tau and \delta are both continuously differentiable functions on [t_{0},\infty) with 0\leq\delta(t)\leq\tau(t)\leq D for some positive constant D , such that \delta^{\prime}(t)<1 for all t\geq t_{0} . Then for any t\geq t_{0} , there exists a unique c(t)\in[0,D] such that c(t)=\tau(t)-\delta(t-c(t)). (2.1) Moreover, the function t\mapsto c(t) is differentiable and fulfills 1-c^{\prime}(t)=\frac{1-\tau^{\prime}(t)}{1-\delta^{\prime}(t-c(t))}. (2.2) Proof. Define the following continuous function F\colon\mathbb{R}\times[t_{0},\infty)\to\mathbb{R},\qquad F(c,t)\coloneqq c-\tau(t)+\delta(t-c). (2.3) Since F(0,t)\leq 0 , F(D,t)\geq 0 , and \frac{\partial F}{\partial c}(c,t)=1-\delta^{\prime}(t)>0, hence for any t\geq t_{0} , there exists a unique c\in[0,D] satisfying ( 2.1 ). Applying the implicit function theorem, we obtain that c is differentiable and F(c(t),t)=0 . Differentiating this with respect to t yields ( 2.2 ). ∎ Notation Recall that N_{p}\geq N_{q} . For the convenience of notation, we define Q_{i}\equiv 0,\ \delta_{i}\equiv 0 , for all N_{q}T_{0}:\ \sum\limits_{i=1}^{N_{p}}\bar{P}_{i}(\xi)>0, (2.6) \sum_{i=1}^{N_{r}}R_{i}(t)+\sum_{i=1}^{N_{q}}\int_{t-c_{i}(t)}^{t}Q_{i}(s)\,ds\leq 1. (2.7) Define y(t)\coloneqq x(t)-\sum_{i=1}^{N_{r}}R_{i}(t)x(t-r_{i}(t))-\sum_{i=1}^{N_{q}}\int_{t-c_{i}(t)}^{t}Q_{i}(s)x(s-\delta_{i}(s))\,ds. (2.8) Then y^{\prime}(t)\leq 0 and x(t)\geq y(t)>0 . The proof is similar to that of [ 1 , Lemma 3.3.1] . Proof. Assume that x(t) is a solution of ( 1.1 ) such that x(t)>0 for all t\geq T\geq t_{0}+D . Differentiating y(t) yields \displaystyle y^{\prime}(t)={} \displaystyle\biggl[x(t)-\sum_{i=1}^{N_{r}}R_{i}(t)x(t-r_{i}(t))\biggr]^{\prime}-\sum_{i=1}^{N_{q}}Q_{i}(t)x(t-\delta_{i}(t)) \displaystyle{}+\sum_{i=1}^{N_{q}}Q_{i}(t-c_{i}(t))x\Big(t-c_{i}(t)-\delta_{i}(t-c_{i}(t))\Big)(1-c_{i}^{\prime}(t)). In view of equation ( 1.1 ) and the definition of \bar{P}_{i} and c_{i} we obtain y^{\prime}(t)=-\sum_{i=1}^{N_{p}}\Big[P_{i}(t)-Q_{i}(t-c_{i}(t))(1-c_{i}^{\prime}(t))\Big]x(t-\tau_{i}(t)), or equivalently y^{\prime}(t)=-\sum_{i=1}^{N_{p}}\bar{P}_{i}(t)x(t-\tau_{i}(t)). (2.9) Therefore y^{\prime}(t)\leq 0 for t\geq T . It remains to prove the positivity of y(t) . In view of ( 2.6 ), ( 2.9 ) and the positivity of x(t) , it suffices to prove that y(t) is nonnegative. We distinguish two cases. Case 1. Suppose that x(t) is unbounded. Then there exists a sequence (s_{n})_{n\in\mathbb{N}} with s_{n}\to\infty , x(s_{n})\to\infty such that x(t)\leq x(s_{n})\quad\text{for all }t\leq s_{n}. Hence, \displaystyle y(s_{n}) \displaystyle=x(s_{n})-\sum_{i=1}^{N_{r}}R_{i}(s_{n})x(s_{n}-r_{i}(s_{n}))-\sum_{i=1}^{N_{q}}\int_{s_{n}-c_{i}(s_{n})}^{s_{n}}Q_{i}(s)x(s-\delta_{i}(s))\,ds \displaystyle\geq x(s_{n})\left(1-\sum_{i=1}^{N_{r}}R_{i}(s_{n})-\sum_{i=1}^{N_{q}}\int_{s_{n}-c_{i}(s_{n})}^{s_{n}}Q_{i}(s)\,ds\right)\geq 0 for all n by property ( 2.7 ). By the monotonicity of y , we conclude that y(t)\geq 0 . Case 2. Suppose that x(t) is bounded. Let M_{x}\coloneqq\limsup_{t\to\infty}x(t)\quad\text{and}\quad M_{y}\coloneqq\lim_{t\to\infty}y(t)\in\mathbb{R}\cup\{-\infty\}. Then there exists a sequence (s_{n})_{n\in\mathbb{N}} with s_{n}\to\infty and x(s_{n})\to M_{x} . Moreover, let (\xi_{n})_{n\in\mathbb{N}} be such that x(\xi_{n})\coloneqq\max\{x(s):s_{n}-2D\leq s\leq s_{n}\}. Clearly, \xi_{n}\to\infty , and, by passing to a convergent subsequence if necessary (without changing notation), we may assume \lim_{n\to\infty}x(\xi_{n})\leq M_{x}. Now, consider the difference x(s_{n})-y(s_{n})=\sum_{i=1}^{N_{r}}R_{i}(s_{n})x(s_{n}-r_{i}(s_{n}))+\sum_{i=1}^{N_{q}}\int_{s_{n}-c_{i}(s_{n})}^{s_{n}}Q_{i}(s)x(s-\delta_{i}(s))\,ds. By the definition of \xi_{n} , we have x(s_{n})-y(s_{n})\leq x(\xi_{n})\left(\sum_{i=1}^{N_{r}}R_{i}(s_{n})+\sum_{i=1}^{N_{q}}\int_{s_{n}-c_{i}(s_{n})}^{s_{n}}Q_{i}(s)\,ds\right). Using inequality ( 2.7 ) we obtain x(s_{n})-y(s_{n})\leq x(\xi_{n}),\quad\ \text{for all }n\in\mathbb{N}. Letting n\to\infty yields M_{x}-M_{y}\leq M_{x} , and thus M_{y}\geq 0. Since y^{\prime}(t)\leq 0 , it follows that y(t)\geq M_{y}\geq 0,\text{ for }t\geq T . ∎ 3 Results – variable delays First, we consider the most general case, where all the delay functions in ( 1.1 ) are time-dependent and satisfy the properties listed in the introduction. Now we will connect our equation to a simple retarded delayed differential equation. This provides a generalization of [ 16 , Theorem 6.9.1] . Theorem 3.1 . Let \bar{P}^{\ast}_{i}(t) and c^{*}_{i}(t) be arbitrary such that 0\leq\bar{P}^{\ast}_{i}(t)\leq\bar{P}_{i}(t) and c^{*}_{i}(t)\geq c_{i}(t) for all t\in[t_{0}+D,\infty) and 1\leq i\leq N_{p} . Furthermore, assume that they satisfy \displaystyle\forall T_{0}\in{(t_{0},\infty)}\;\exists\xi>T_{0}:\ \sum\limits_{i=1}^{N_{p}}\bar{P}^{\ast}_{i}(\xi) \displaystyle>0, (3.1) and \displaystyle\sum_{i=1}^{N_{r}}R_{i}(t)+\sum_{i=1}^{N_{q}}\int_{t-c^{*}_{i}(t)}^{t}Q_{i}(s)\,ds \displaystyle\leq 1. (3.2) If every solution of x^{\prime}(t)+\sum_{i=1}^{N_{p}}\bar{P}^{\ast}_{i}(t)x(t-\tau_{i}(t))=0 (3.3) oscillates, then every solution of equation ( 1.1 ) also oscillates. Proof. Assume that every solution of ( 3.3 ) oscillates. For the sake of contradiction, suppose that ( 1.1 ) has an eventually positive solution x . Notice that conditions ( 3.1 )–( 3.2 ) imply that inequalities ( 2.6 ) and ( 2.7 ) hold. Define y by ( 2.8 ). As seen in the proof of Lemma 2.2 , y^{\prime}(t)+\sum_{i=1}^{N_{p}}\bar{P}_{i}(t)x(t-\tau_{i}(t))=0. Using the fact that y(t)\leq x(t) , and \bar{P}^{\ast}_{i}(t)\leq\bar{P}_{i}(t) we obtain y^{\prime}(t)+\sum_{i=1}^{N_{p}}\bar{P}^{\ast}_{i}(t)y(t-\tau_{i}(t))\leq 0, where y is eventually positive by Lemma 2.2 . According to [ 16 , Corollary 3.2.2] , this inequality implies that there exists an eventually positive solution z of z^{\prime}(t)+\sum_{i=1}^{N_{p}}\bar{P}^{\ast}_{i}(t)z(t-\tau_{i}(t))=0. This is a contradiction that completes our proof. ∎ Since, in general, we do not have explicit formulas for the functions c_{i} , a direct application of the previous theorem and corollaries might be difficult. Therefore, we provide further sufficient conditions. Let us recall that D is an upper bound for all delays. Corollary 3.2 . Assume that there exists \Delta_{j} such as \delta^{\prime}_{j}(t)\leq\Delta_{j}<1 and \tau^{\prime}_{i}(t)<1 for all t\geq t_{0},\ j\in\{1,\dots N_{q}\},\ i\in\{1,\dots,N_{p}\} . Let \bar{P}^{\ast}_{i}(t)\coloneqq P_{i}(t)-\frac{1-\tau^{\prime}_{i}(t)}{1-\Delta_{i}}\sup_{s\in[t-D,t]}Q_{i}(s)>0,\qquad(1\leq i\leq N_{p}) and \sum_{i=1}^{N_{r}}R_{i}(t)+\sum_{i=1}^{N_{q}}\int_{t-D}^{t}Q_{i}(s)\,ds\leq 1, (3.4) moreover, assume that every solution of x^{\prime}(t)+\sum_{i=1}^{N_{p}}\bar{P}^{\ast}_{i}(t)x(t-\tau_{i}(t))=0 oscillates. Then every solution of ( 1.1 ) oscillates. Proof. Inequality ( 3.4 ) implies ( 3.2 ), while condition ( 3.1 ) comes from the inequalities \displaystyle 0\frac{1}{e}, ( A_{1} ) \liminf_{t\to\infty}\sum_{i=1}^{N_{p}}\bar{P}^{\ast}_{i}(t)\tau_{i}(t)>\frac{1}{e}, ( B_{1} ) \liminf_{t\to\infty}\int_{t-\tau_{\min}(t)}^{t}\sum_{i=1}^{N_{p}}\bar{P}^{\ast}_{i}(s)\,ds>\frac{1}{e}, ( C_{1} ) \limsup_{t\to\infty}\int_{t-\tau_{\min}(t)}^{t}\sum_{i=1}^{N_{p}}\bar{P}^{\ast}_{i}(s)\,ds>1. ( D_{1} ) Proof. According to [ 24 , Corollary 1] , [ 9 , Corollary 2.6.2] , [ 23 , Theorem 2.7.1] , and [ 23 , Remark 2.7.3] , respectively, each of conditions ( A 1 ), ( B 1 ), ( C 1 ), and ( D 1 ) ensures that every solution of ( 3.3 ) oscillates. Hence, by Theorem 3.1 , it follows that every solution of ( 1.1 ) oscillates. ∎ The next corollary uses the so-called slowly varying property. A function f is slowly varying at infinity (additively) if, for all h\in\mathbb{R} , the expression |f(t+h)-f(t)| tends to zero as t goes to infinity. We recall a characterization of continuous slowly varying functions given by [ 30 , p. 30] : a continuous function f is slowly varying at infinity if and only if it can be decomposed into a sum of two functions, where one of them is continuous and converges to a finite value, and the other one is differentiable with its derivative tending to zero as t\to\infty . Corollary 3.4 . Assume that there exist functions 0\leq\bar{P}^{\ast}_{i}\leq\bar{P}_{i} and c^{*}_{i}\geq c_{i} that conditions ( 3.1 )–( 3.2 ) hold. Moreover, suppose that \liminf\limits_{t\to\infty}\int_{t-\tau_{i}(t)}^{t}\bar{P}^{\ast}_{i}(s)\,ds>0 , 1\leq i\leq N_{p} , \bar{P}^{\ast}_{i}(t) , and \tau_{i}(t) are all uniformly continuous and bounded for all i\in\{1,2,\dots,N_{p}\} . If any of the following conditions is fulfilled, then every solution of equation ( 1.1 ) oscillates. (a) \tau_{i}(t)\equiv\tau_{i} are positive constants, \sum_{i=1}^{N_{p}}\int_{t-\tau_{i}}^{t}\bar{P}^{\ast}_{i}(s)\,ds is slowly varying at infinity and \limsup_{t\to\infty}\sum_{i=1}^{N_{p}}\int_{t-\tau_{i}}^{t}\bar{P}^{\ast}_{i}(s)\,ds>\frac{1}{e}. ( A_{2} ) (b) \sum_{i=1}^{N_{p}}\bar{P}^{\ast}_{i}(t)\tau_{i}(t) is slowly varying at infinity and \limsup_{t\to\infty}\sum_{i=1}^{N_{p}}\bar{P}^{\ast}_{i}(t)\tau_{i}(t)>\frac{1}{e}. ( B_{2} ) (c) There exists a uniformly continuous function \hat{\tau}\colon[t_{0},\infty)\to[0,\infty) such that 0\leq\hat{\tau}(t)\leq\tau_{i}(t) for all t\geq t_{0} and 1\leq i\leq N_{p} , and \int_{t-\hat{\tau}(t)}^{t}\sum_{i=1}^{N_{p}}\bar{P}^{\ast}_{i}(s)\,ds is slowly varying at infinity, and \limsup_{t\to\infty}\int_{t-\hat{\tau}(t)}^{t}\sum_{i=1}^{N_{p}}\bar{P}^{\ast}_{i}(s)\,ds>\frac{1}{e}. ( C_{2} ) Proof. According to [ 13 , Theorem 2.2. (a)/(b)/(c)] , either ( A 2 ), ( B 2 ), or ( C 2 ) implies that every solution of ( 3.3 ) oscillates, thus applying Theorem 3.1 we obtain that every solution of ( 1.1 ) oscillates. ∎ There are special cases in which the functions c_{i}(t) admit explicit closed-form expressions – see Example 5.1 . In particular, this is the case, if the functions \delta_{i} are all constants. Indeed, if the delay functions \delta_{i} are all constant, then c_{i}(t) can be expressed explicitly as c_{i}(t)=\tau_{i}(t)-\delta_{i} and ( 2.5 ) simplifies to \bar{P}_{i}(t)\coloneqq P_{i}(t)-Q_{i}(t-\tau_{i}(t)+\delta_{i})(1-\tau_{i}^{\prime}(t)),\qquad t\in[t_{0}+D,\infty). (3.5) Our results so far can be sharpened as summed up in the following theorem. Theorem 3.5 . If the delays \delta_{k} are constant for 1\leq k\leq N_{q} , then Theorems 3.1 , 3.3 and 3.4 hold true with the choice \bar{P}^{*}_{i}(t)\coloneqq P_{i}(t)-Q_{i}(t-\tau_{i}(t)+\delta_{i})(1-\tau_{i}^{\prime}(t))\quad\text{and}\quad c_{i}^{*}(t)\coloneqq\tau_{i}(t)-\delta_{i}. 4 Results – constant delays This section is devoted to the special case when all the delays in ( 1.1 ) are constant, that is, we deal with the NDDE \bigg[x(t)-\sum_{i=1}^{N_{r}}R_{i}(t)x(t-r_{i})\bigg]^{\prime}+\sum_{i=1}^{N_{p}}P_{i}(t)x(t-\tau_{i})-\sum_{i=1}^{N_{q}}Q_{i}(t)x(t-\delta_{i})=0, (4.1) for t\geq t_{0} , where N_{q}\leq N_{p} , the coefficient functions R_{i},\ P_{j}, and Q_{k} are nonnegative and continuous for all 1\leq i\leq N_{r} , 1\leq j\leq N_{p} and 1\leq k\leq N_{q} , while R_{i}\in C^{1} , and the constant delays r_{i},\ \tau_{j},\ \delta_{k} are nonnegative, bounded above by D and fulfill \delta_{k}\leq\tau_{k} for all 1\leq k\leq N_{q} . In this case ( 3.5 ) simplifies to \bar{P}_{i}(t)=P_{i}(t)-Q_{i}(t-\tau_{i}+\delta_{i}) . The following comparison theorem generalizes [ 1 , Theorem 3.4.1] by allowing multiple delays. It also serves as a basis to our second main result, Theorem 4.2 . Theorem 4.1 . Assume that \tau_{i}>0 for all 1\leq i\leq N_{p} , moreover, \displaystyle\bar{P}_{i}(t)=P_{i}(t)-Q_{i}(t-\tau_{i}+\delta_{i}) \displaystyle\geq 0, \displaystyle\sum_{i=1}^{N_{r}}R_{i}(t)+\sum_{i=1}^{N_{q}}\int_{t-\tau_{i}+\delta_{i}}^{t}Q_{i}(s)\,ds \displaystyle\leq 1, and either \displaystyle\sum_{i=1}^{N_{r}}R_{i}(t)+\sum_{i=1}^{N_{p}}\bar{P}_{i}(t) \displaystyle>0 (4.2) or \displaystyle\sum_{i=1}^{N_{p}}\int_{t-\tau_{i}}^{t}\bar{P}_{i}(s)\,ds \displaystyle>0 (4.3) for all t\geq t_{0}+D . Then every solution of equation ( 4.1 ) oscillates if and only if every solution of the corresponding differential inequality \bigg[x(t)-\sum_{i=1}^{N_{r}}R_{i}(t)x(t-r_{i})\bigg]^{\prime}+\sum_{i=1}^{N_{p}}P_{i}(t)x(t-\tau_{i})-\sum_{i=1}^{N_{q}}Q_{i}(t)x(t-\delta_{i})\leq 0 (4.4) oscillates. Proof. The proof follows the main ideas of [ 1 , Theorem 3.4.1] . For the sake for completeness, we provide the details here. The sufficiency part is trivial. We will prove the necessity. Suppose that x is an eventually positive solution of ( 4.4 ). Define y as in Lemma 2.2 using constant delays. A similar argument to the one presented there shows y^{\prime}(t)\leq-\sum_{i=1}^{N_{p}}\bar{P}_{i}(t)x(t-\tau_{i}) for sufficiently large t . Integrating on [t,\infty) yields y(t)\geq\int_{t}^{\infty}\sum_{i=1}^{N_{p}}\bar{P}_{i}(s)x(s-\tau_{i})\,ds. From the definition of y we obtain that x(t)\geq\sum_{i=1}^{N_{r}}R_{i}(t)x(t-r_{i})+\sum_{i=1}^{N_{q}}\int_{t-\tau_{i}+\delta_{i}}^{t}\!\!Q_{i}(s)x(s-\delta_{i})\ ds+\int_{t}^{\infty}\sum_{i=1}^{N_{p}}\bar{P}_{i}(s)x(s-\tau_{i})\,ds (4.5) for t\geq t_{1} for a sufficiently large t_{1} . Now, let E\coloneqq\{u\in C([t_{1}-D,\infty),\mathbb{R}^{+}):0\leq u(t)\leq 1,\text{ for all }t\geq t_{1}-D\}. Define the operator F by (Fu)(t)\!=\!\left\{\begin{aligned} &\frac{1}{x(t)}\Bigg[\sum_{i=1}^{N_{r}}R_{i}(t)u(t-r_{i})x(t-r_{i})\\ &\begin{aligned} &\hphantom{\frac{1}{x(t)}\Bigg[}+\sum_{i=1}^{N_{q}}\int_{t-\tau_{i}+\delta_{i}}^{t}Q_{i}(s)u(s-\delta_{i})x(s-\delta_{i})\,ds\\ &\hphantom{\frac{1}{x(t)}\Bigg[}+\int_{t}^{\infty}\sum\limits_{i=1}^{N_{p}}\bar{P}_{i}(s)u(s-\tau_{i})x(s-\tau_{i})\,ds\Bigg],&&t_{1}\leq t,\\[10.0pt] &\frac{t-t_{1}+D}{D}(Fu)(t_{1})+1-\frac{t-t_{1}+D}{D},&&t_{1}-D\leq t0 , for all t\in[t_{1}-D,t_{1}) . Define the sequence (u_{k})_{k\in\mathbb{N}} , where u_{0}\equiv 1 , u_{k+1}=Fu_{k} . By mathematical induction we infer from ( 4.5 ) that 0\leq u_{k+1}\leq u_{k}\leq 1 , for all k\in\mathbb{N} . Defining u(t)\coloneqq\lim_{k\to\infty}u_{k}(t) , and using Lebesgue’s dominated convergence theorem, we obtain that for t\geq t_{1} \displaystyle u(t) \displaystyle=\frac{1}{x(t)}\Bigg[\sum_{i=1}^{N_{r}}R_{i}(t)\,u(t-r_{i})\,x(t-r_{i}) \displaystyle\mathrel{\hphantom{=}}\hphantom{\frac{1}{x(t)}\Bigg[}+\sum_{i=1}^{N_{q}}\int_{t-\tau_{i}+\delta_{i}}^{t}Q_{i}(s)\,u(s-\delta_{i})\,x(s-\delta_{i})\,ds \displaystyle\mathrel{\hphantom{=}}\hphantom{\frac{1}{x(t)}\Bigg[}+\int_{t}^{\infty}\sum_{i=1}^{N_{p}}\bar{P}_{i}(s)\,u(s-\tau_{i})\,x(s-\tau_{i})\,ds\Bigg], and u(t)=\frac{t-t_{1}+D}{D}u(t_{1})+1-\frac{t-t_{1}+D}{D}>0 for t_{1}-D\leq t0 for t_{1}-D\leq t0 , for all t\in[t_{1}-D,\infty) . Assume that there exists a t^{*}\geq t_{1} such that w(t)>0 for t_{1}-D\leq t0 for all 1\leq i\leq N_{p} , and that there exists a constant \omega\in(0,\infty) , such that \omega\leq\min_{i,j}\left\{\inf_{s\geq t_{0}+D}\Omega_{i,j}(s)\right\}\quad\text{and}\quad\omega<\frac{1}{N_{r}}. (4.8) Furthermore, assume \sum_{i=1}^{N_{r}}R_{i}(t)+\sum_{i=1}^{N_{q}}\int_{t-\tau_{i}+\delta_{i}}^{t}Q_{i}(s)\,ds\leq 1 (4.9) for all t\geq t_{0}+D , 1\leq i\leq N_{p} . If there exists m\in\mathbb{N} such that \displaystyle\frac{(N_{r}\omega)^{m}}{1-N_{r}\omega}\liminf_{t\to\infty}\sum_{i=1}^{N_{p}}\int_{t-mr_{0}-\tau_{i}}^{t}\bar{P}_{i}(s)\,ds \displaystyle>\frac{1}{e}, ( A_{3} ) or \displaystyle\frac{(N_{r}\omega)^{m}}{1-N_{r}\omega}\liminf_{t\to\infty}\sum_{i=1}^{N_{p}}(mr_{0}+\tau_{i})\bar{P}_{i}(t) \displaystyle>\frac{1}{e}, ( B_{3} ) or \displaystyle\frac{(N_{r}\omega)^{m}}{1-N_{r}\omega}\limsup_{t\to\infty}\int_{t-mr_{0}-\tau_{0}}^{t}\sum_{i=1}^{N_{p}}\bar{P}_{i}(s)\,ds \displaystyle>1, ( D_{3} ) then every solution of equation ( 4.1 ) oscillates. This theorem generalizes and improves [ 1 , Theorem 3.4.9] which only considered the case N_{r}=N_{p}=N_{q}=1 and m=1 . Observe that if all the delays are constant, \bar{P}_{i}(t)>0 and \omega>0 satisfies ( 4.8 ), then Theorem 4.2 – already with the choice m=0 – gives more efficient conditions than Corollary 3.3 (cf. Examples 5.5 and 5.4 ). Proof. The proof follows a similar argument as that of [ 1 , Theorem 3.4.9] . Let m\in\mathbb{N} be such that one of inequalities ( A 3 ), ( B 3 ) and ( D 3 ) holds. For the sake of contradiction, suppose that ( 4.1 ) has an eventually positive solution x . Let y be defined by ( 2.8 ) – recall that here c_{i}=\tau_{i}-\delta_{i} , with \delta_{i}=0 for N_{q}m and t is sufficiently large. By virtue of Lemma 2.2 we have that z^{\prime}(t)\leq 0 and z(t)>0 . Using the monotonicity of z and the positivity of \tilde{y} , we obtain the inequality \tilde{y}(t)\geq\sum_{i=0}^{K}(N_{r}\omega)^{i}z(t-ir_{0}). Now, omitting the first m terms, and using the monotonicity again, and N_{r}\omega\in(0,1) , we infer \tilde{y}(t)\geq(N_{r}\omega)^{m}\frac{1-(N_{r}\omega)^{K-m+1}}{1-N_{r}\omega}z(t-mr_{0}). Combining this with equation ( 4.11 ) we conclude that z(t) is an eventually positive solution of the differential inequality z^{\prime}(t)+\sum_{i=1}^{N_{p}}\tilde{p}_{i}(t)z(t-mr_{0}-\tau_{i})\leq 0, where \tilde{p}_{i}(t)\coloneqq(N_{r}\omega)^{m}\frac{1-(N_{r}\omega)^{K-m+1}}{1-N_{r}\omega}\bar{P}_{i}(t). Applying [ 16 , Corollary 3.2.2] we obtain that the differential equation v^{\prime}(t)+\sum_{i=1}^{N_{p}}\tilde{p}_{i}(t)v(t-mr_{0}-\tau_{i})=0 (4.12) also has an eventually positive solution. On the one hand, for large enough K , ( A 3 ), ( B 3 ) and ( D 3 ) imply \displaystyle\liminf_{t\to\infty}\sum_{i=1}^{N_{p}}\int_{t-mr_{0}-\tau_{i}}^{t}\tilde{p}_{i}(t)\,ds \displaystyle>\frac{1}{e}, (4.13) \displaystyle\liminf_{t\to\infty}\sum_{i=1}^{N_{p}}(mr_{0}+\tau_{i})\tilde{p}_{i}(t) \displaystyle>\frac{1}{e}, (4.14) and \displaystyle\limsup_{t\to\infty}\int_{t-mr_{0}-\tau_{0}}^{t}\sum_{i=1}^{N_{p}}\tilde{p}_{i}(t)\,ds \displaystyle>1, (4.15) respectively. On the other hand, it is well-known, that each of the inequalities ( 4.13 ), ( 4.14 ) and ( 4.15 ) is sufficient for the oscillation of all solutions of equation ( 4.12 ) (see [ 24 , Corollary 1] , [ 9 , Corollary 2.6.2] and [ 23 , Remark 2.7.3] , respectively). This is a contradiction that completes our proof. ∎ It is worth noting that the conditions in the previous theorem can be significantly sharpened by replacing the limit inferior with limit superior, provided the coefficient functions fulfill an extra condition related to the slowly varying property. We conclude our results by formulating this in the next corollary. Corollary 4.3 . Assume that inequalities ( 4.8 )–( 4.9 ) hold, the functions \bar{P}_{i} are bounded, uniformly continuous, and satisfy \liminf_{t\to\infty}\int_{t-\tau_{i}}^{t}\bar{P}_{i}(s)\,ds>0 for all 1\leq i\leq N_{p} . For m\in\mathbb{N} , define \displaystyle A_{m}(t) \displaystyle\coloneqq\frac{(N_{r}\omega)^{m}}{1-N_{r}\omega}\sum_{i=1}^{N_{p}}\int_{t-mr_{0}-\tau_{i}}^{t}\bar{P}_{i}(s)\,ds, and \displaystyle B_{m}(t) \displaystyle\coloneqq\frac{(N_{r}\omega)^{m}}{1-N_{r}\omega}\sum_{i=1}^{N_{p}}\bar{P}_{i}(t)(mr_{0}+\tau_{i}). If one of the following conditions is fulfilled, then every solution of equation ( 4.1 ) oscillates: \text{$A_{m}$ is slowly varying at infinity, and}\quad\limsup_{t\to\infty}A_{m}(t)>\frac{1}{e}\quad\text{for some $m\in\mathbb{N}$}; \text{$B_{m}$ is slowly varying at infinity, and}\quad\limsup_{t\to\infty}B_{m}(t)>\frac{1}{e}\quad\text{for some $m\in\mathbb{N}$}. Proof. The indirect proof proceeds in a similar way as in Theorem 4.2 . The only difference is that we apply [ 13 , Theorem 2.2 (a)–(b)] at the end to arrive at a contradiction. ∎ 5 Examples In this section we give several examples illustrating the applicability of our results. In our first example, Theorem 3.1 can be applied directly. Example 5.1 . Consider the equation \biggl[x(t)-\frac{1}{4}x\Bigr(t-\frac{1}{2}\Bigr)\biggr]^{\prime}+2x\biggl(t-\Bigl(\frac{1}{2}\cos t+1+e^{-t+1+\frac{1}{2}\cos t}\Bigr)\biggr)-e^{-t}x\bigl(t-e^{-t}\bigr)=0 for t\geq 0 . Here we take N_{r}=N_{p}=N_{q}=1 , R_{1}(t)=\frac{1}{4} , P_{1}(t)=2 , Q_{1}(t)=e^{-t} , r_{1}(t)=\frac{1}{2} , \tau_{1}(t)=\frac{1}{2}\cos t+1+e^{-t+1+\frac{1}{2}\cos t} , and \delta_{1}(t)=e^{-t} . The associated implicit relation yields the explicit solution c_{1}(t)=\tfrac{1}{2}\cos t+1 . One can show that \bar{P}_{1}(t)=2-e^{c_{1}(t)-t}(1-c_{1}^{\prime}(t)) , which tends to 2 as t\to\infty . Hence, there exists T_{0}>0 such that \bar{P}_{1}(t)\geq 1 for all t\geq T_{0} . Applying Corollary 3.3 , we obtain \liminf_{t\to\infty}\int_{t-\tau_{1}(t)}^{t}\bar{P}_{1}(s)\,ds\geq\liminf_{t\to\infty}\tau_{1}(t)\geq\liminf_{t\to\infty}c_{1}(t)=\frac{1}{2}>\frac{1}{e}, since c_{1}(t)\leq\tau_{1}(t) . Therefore, every solution oscillates. The next example illustrates the applicability of Theorem 4.2 in case of constant delays, and the importance of the appropriate choice of m . Example 5.2 . Consider the NDDE \bigg[x(t)-\frac{1}{2}x\Bigl(t-\frac{\pi}{2}\Bigr)\bigg]^{\prime}+\bigg(\frac{3}{4}\sin(4t)+\frac{5}{4}\bigg)x(t-0.1)+\frac{1}{10}x(t-0.01)-\frac{1}{4}x(t-0.05)=0 for t\geq 0 . This equation is of the form ( 4.1 ) with N_{r}=N_{q}=1 , N_{p}=2 , R_{1}(t)=\frac{1}{2},\ P_{1}(t)=\frac{3}{4}\sin(4t)+\frac{5}{4} , P_{2}(t)=\frac{1}{10} , Q_{1}(t)=\frac{1}{4} , r_{1}=\frac{\pi}{2} , \tau_{1}=0.1 , \tau_{2}=0.01 and \delta_{1}=0.05 , from which we obtain \bar{P}_{1}(t)=\frac{3}{4}\sin(4t)+1,\ \bar{P}_{2}(t)=\frac{1}{10} and \Omega_{1,1}(t)=\Omega_{2,1}(t)=\frac{1}{2} . A suitable choice for \omega is \frac{1}{2} . Then the left-hand side of condition ( B 3 ) reads 2^{1-m}\bigg[\Bigl(\frac{m\pi}{2}+0.1\Bigr)\frac{1}{4}+\Bigl(\frac{m\pi}{2}+0.01\Bigr)\frac{1}{10}\bigg]. Choosing m=2 , we have \frac{1}{2}\bigg(\frac{\pi+0.1}{4}+\frac{\pi+0.01}{10}\bigg)\approx 0.563>\frac{1}{e}. Thus, condition ( B 3 ) is satisfied, and hence every solution oscillates. We note that one could also choose m=1 or m=3 to draw the same conclusion, while other values of m do not allow the oscillatory behavior to be determined via ( B 3 ). For comparison, condition ( D 3 ) provides a sufficient condition for oscillation with m\in\{1,2,3\} , as well, while ( A 3 ) guarantees oscillation if we set 1\leq m\leq 5 . In the following example we take advantage of the slowly varying property of the coefficient functions. Example 5.3 . Consider the equation \displaystyle\bigg[x(t)-\frac{1}{2}x(t-0.15)\bigg]^{\prime}+1.1x(t-0.05) \displaystyle\qquad+(2\cos(\ln t)+2.05)x(t-0.01)-x(t-0.02)=0, for t\geq 1 . This NDDE is of the form ( 4.1 ) with N_{r}=N_{q}=1 , N_{p}=2 , R_{1}(t)=\frac{1}{2} , P_{1}(t)=1.1 , P_{2}(t)=\bar{P}_{2}(t)=2\cos(\ln t)+2.05 , \ Q_{1}(t)=1 , r_{1}=0.15,\ \tau_{1}=0.05,\ \tau_{2}=0.01,\ \delta_{1}=0.02, \bar{P}_{1}(t)=0.1 . Furthermore, \Omega_{1,1}(t)=\frac{1}{2} and \Omega_{2,1}(t)\to\frac{1}{2} as t\to\infty , therefore we can choose \omega to be smaller than, but arbitrarily close to \frac{1}{2} . Even with the choice \omega=\frac{1}{2} , inequality ( B 3 ) is not fulfilled for any m\in\mathbb{N} , as \frac{(\frac{1}{2})^{m}}{1-\frac{1}{2}}\liminf_{t\to\infty}\bigg[0.1(0.15m+0.05)+\left(2\cos(\ln(t))+2.05\right)(0.15m+0.01)\bigg]\\ =2^{1-m}(0.0225m+0.0055)<\frac{1}{e}\quad\text{for all }m\in\mathbb{N}. Notice that thanks to the slow variance of the functions \bar{P}_{1} and \bar{P}_{2} , conditions ( A 3 ) and ( B 3 ) are equivalent, so ( A 3 ) cannot guarantee the oscillation of all solutions, either. On the other hand, these functions have positive lower bounds, so we may try to apply Corollary 4.3 . With \omega=\frac{1}{2} and m=1 , one would have \limsup_{t\to\infty}B_{1}(t)=0.668>\frac{1}{e} Hence, by choosing \omega slightly below \frac{1}{2} , we may conclude that every solution oscillates. The same conclusion follows if one chooses m=2 or m=3 . However, a similar simple calculation shows that \limsup_{t\to\infty}B_{m}(t)<1 for all m\in\mathbb{N} and all \omega<\frac{1}{2} , thus ( D 3 ) is not satisfied, and Theorem 4.2 cannot be applied here. Finally, one might wonder whether Corollary 3.3 or Theorem 4.2 is stronger for constant delays, but there is no clear answer to that. The following two examples will illustrate this fact. Example 5.4 . Consider the equation \bigg[x(t)-e^{-t}x(t-2\pi)\bigg]^{\prime}+3x(t-1)+(\sin t+1.5)x(t-2)-x\biggl(t-\frac{1}{2}\biggr)=0,\qquad t\geq 1 with constant delays. This equation is in the form of ( 4.1 ) with N_{r}=N_{q}=1 , N_{p}=2 , R_{1}(t)=e^{-t},\ P_{1}(t)=3,\ P_{2}(t)=\sin(t)+1.5 , Q_{1}(t)=1 , r_{1}=2\pi,\ \tau_{1}=1,\ \tau_{2}=2,\ \delta_{1}=\frac{1}{2} , from which \bar{P}_{1}(t)=2,\ \bar{P}_{2}(t)=\sin(t)+1.5 . On the one hand, \Omega_{1,1}(t)=\Omega_{2,1}(t)=e^{-t}\to 0 as t goes to infinity, so we cannot apply Theorem 4.2 . On the other hand, if we let \bar{P}^{\ast}_{i}=\bar{P}_{i} and c^{*}_{i}=\tau_{i}-\delta_{i} (i=1,2) , then all the conditions of Corollary 3.3 are satisfied. Hence, the left-hand side of the inequality ( A 1 ) takes the following form: \liminf_{t\to\infty}\bigg(\int_{t-1}^{t}2\,ds+\int_{t-2}^{t}\sin(s)+1.5\,ds\bigg)>3, which is clearly larger than \frac{1}{e} , hence Corollary 3.3 implies that every solution oscillates. Example 5.5 . Let us consider the NDDE \bigg[x(t)-\frac{1}{3}x(t-2)\bigg]^{\prime}+(\cos 2t+2)x\biggl(t-\frac{1}{2}\biggr)-\biggl[\cos\biggl(2t+\frac{1}{2}\biggr)+\frac{3}{2}\biggr]x\biggl(t-\frac{1}{4}\biggr)=0, which is of the form ( 4.1 ) with N_{r}=N_{p}=N_{q}=1 , R_{1}(t)=\frac{1}{3},\ P_{1}(t)=\cos 2t+2,\ Q_{1}(t)=\cos(2t+\frac{1}{2})+\frac{3}{2}, r_{1}=2,\ \tau_{1}=\frac{1}{2},\ \delta_{1}=\frac{1}{4},\ c_{1}=\frac{1}{4},\ \bar{P}_{1}(t)=\frac{1}{2},\ \Omega_{1,1}(t)=\frac{1}{3} , thus \omega\coloneqq\frac{1}{3} . Substituting into the left-hand side of any of the conditions in ( A 1 )–( D 1 ) in Corollary 3.3 , we obtain \frac{1}{4} , which is smaller than \frac{1}{e} . Thus, Corollary 3.3 does not imply oscillatory behavior. However, condition ( A 3 ) in Theorem 4.2 is fulfilled with m=2 (it boils down to 0.375>\frac{1}{e}\approx 0.368 ), thus Theorem 4.2 implies that every solution oscillates.