Abstract
Periodic hematological diseases such as cyclical neutropenia or cyclical thrombocytopenia, with their characteristic oscillations of circulating neutrophils or platelets, may pose grave problems for patients. Likewise, periodically administered chemotherapy has the unintended side effect of establishing periodic fluctuations in circulating white cells, red cell precursors and/or platelets. These fluctuations, either spontaneous or induced, often have serious consequences for the patient (e.g. neutropenia, anemia, or thrombocytopenia respectively) which exogenously administered cytokines can partially correct. The question of when and how to administer these drugs is a difficult one for clinicians and not easily answered. In this paper we use a simple model consisting of a delay differential equation with a piecewise linear nonlinearity, that has a periodic solution, to model the effect of a periodic disease or periodic chemotherapy. We then examine the response of this toy model to both single and periodic perturbations, meant to mimic the drug administration, as a function of the drug dose and the duration and frequency of its administration to best determine how to avoid side effects.
Similar content being viewed by others
Notes
Human neutropenia is classified as severe if the neutrophil concentration is below \(0.061\times 10^{9}\,{\mathrm {cells/kg}}\) (of body mass), which corresponds to an absolute neutrophil count (ANC) of \(500\,{\mathrm {cells/}}\upmu {\mathrm {l}}\) (Craig et al. 2015).
References
Bellen A, Zennaro M (2003) Numerical methods for delay differential equations. Oxford University Press, New York
Beuter A, Glass L, Mackey MC, Titcombe MS (2003) Nonlinear dynamics in physiology and medicine. IAM, mathematical biology. Springer, New York
Bodnar M, Piotrowska MJ, Foryś U (2013) Existence and stability of oscillating solutions for a class of delay differential equations. Nonlinear Anal Real World Appl 14(3):1780–1794
Breda D, Van Vleck E (2014) Approximating Lyapunov exponents and Sacker–Sell spectrum for retarded functional differential equations. Numer Math 126:225–257
Brooks G, Langlois GP, Lei J, Mackey MC (2012) Neutrophil dynamics after chemotherapy and G-CSF: the role of pharmacokinetics in shaping the response. J Theor Biol 315:97–109
Colijn C, Mackey MC (2005) A mathematical model of hematopoiesis: I. Periodic chronic mylogenous leukemia. J Theor Biol 237:117–132
Colijn C, Mackey MC (2005) A mathematical model of hematopoiesis: II. Cyclical neutropenia. J Theor Biol 237:133–146
Craig M, Humphries AR, Mackey MC (2016) A mathematical model of granulopoiesis incorporating the negative feedback dynamics and kinetics of G-CSF/neutrophil binding and internalization. Bull Math Biol 78(12):2304–2357
Craig M, Humphries AR, Nekka F, Bélair J, Li J, Mackey MC (2015) Neutrophil dynamics during concurrent chemotherapy and G-CSF administration: mathematical modelling guides dose optimisation to minimise neutropenia. J Theor Biol 385:77–89
De Souza DC, Humphries AR (2018) Dynamics of a mathematical hematopoietic stem-cell population model. arXiv:1712.08308 [math.DS]
Foley C, Mackey MC (2009) Dynamic hematological disease: a review. J Math Biol 58(1–2):285–322
Glass L, Mackey MC (1988) From clocks to chaos: the rhythms of life. Princeton University Press, New Jersey
Glass L, Winfree AT (1984) Discontinuities in phase-resetting experiments. Am J Physiol 246(2 Pt 2):R251–258
Granada A, Hennig R, Ronacher B, Kramer A, Herzel H (2009) Chapter 1 phase response curves: elucidating the dynamics of coupled oscillators. In: Johnson ML, Brand L (eds) Computer methods, part A, methods in enzymology, vol 454. Academic Press, New York, pp 1–27
Grebogi C, Ott E, Yorke JA (1983) Crises, sudden changes in chaotic attractors, and transient chaos. Physica D 7(1–3):181–200
Guevara MR, Glass L (1982) Phase locking, periodic doubling bifurcations and chaos in a mathematical model of a periodically driven oscillator: a theory for the entrainment of biological oscillators. J Math Biol 14:1–23
Hearn T, Haurie C, Mackey MC (1998) Cyclical neutropenia and the peripheral control of white blood cell production. J Theor Biol 192(2):167–181
Kotani K, Yamaguchi I, Ogawa Y, Jimbo Y, Hakao H, Ermentrout G (2012) Adjoint method provides phase response functions for delay-induced oscillations. Phys Rev Lett 109:044101
Krogh-Madsen T, Glass L, Doedel EJ, Guevara MR (2004) Apparent discontinuities in the phase-resetting response of cardiac pacemakers. J Theor Biol 230(4):499–519
Langlois GP, Craig M, Humphries AR, Mackey MC, Mahaffy JM, Bélair J, Moulin T, Sinclair SR, Wang L (2017) Normal and pathological dynamics of platelets in humans. J Math Biol 75(6):1411–1462
Mackey MC, Tyran-Kamińska M, Walther H-O (2017) Response of an oscillatory differential delay equation to a single stimulus. J Math Biol 74(5):1139–1196
Mathworks (2015) MATLAB 2015b. Mathworks, Natick
Nayfeh AH, Balachandran B (2007) Applied nonlinear dynamics. Analytical, computational and experimental methods. WILEY-VCH Verlag GmbH & Co, KGaA, Weinheim
Novicenko V, Pyragas K (2012) Phase reduction of weakly perturbed limit cycle oscillations in time-delay systems. Physica D 241:1090–1098
Pujo-Menjouet L (2016) Blood cell dynamics: half of a century of modelling. Math Model Nat Phenom 11(1):92–115
Shampine LF, Gladwell I, Thompson S (2003) Solving ODEs with MATLAB. Cambridge University Press, New York
Winfree AT (1980) The geometry of biological time. Biomathematics, vol 8. Springer, Berlin
Zhuge C, Lei J, Mackey MC (2012) Neutrophil dynamics in response to chemotherapy and G-CSF. J Theor Biol 293:111–120
Acknowledgements
MCM would like to thank the Institut für Theoretische Neurophysik, Universität Bremen for their hospitality during the time in which much of the writing of this paper took place. We are very grateful to Tony Humphries for fruitful discussions and suggestions.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
DCS was supported by National Council for Scientific and Technological Development of Brazil (CNPq) postdoctoral fellowship 201105/2014-4, and MCM is supported by a Discovery Grant from the Natural Sciences and Engineering Research Council (NSERC) of Canada.
Appendix: Proof of the results
Appendix: Proof of the results
Here we present the proofs of Remarks and Propositions from Sects. 3 and 4.
Proof of Remark 3.1
First it is shown that \(a>\beta _{U}\) holds. For \(\varDelta \leqslant t \leqslant \varDelta +\sigma \) we have
with \(x^{(\varDelta )}(\varDelta ) = -\beta _{U} + \beta _{U}\mathrm {e}^{\tilde{z}_{2}-\varDelta }\) and
From (42) the condition \(x^{(\varDelta )}(\varDelta +\sigma )> 0\) can be written as
but \(x^{(\varDelta )}(\varDelta )=-\beta _{U}(1-\mathrm {e}^{\tilde{z}_{2}-\varDelta })<0\) since \(\varDelta >\tilde{z}_{2}\), then we must have \(a>\beta _{U}\).
For \(\varDelta +\sigma \leqslant t \leqslant \tilde{T}\) we have
Equation (43) gives \(x^{(\varDelta )}(\varDelta +\sigma ) + \beta _{U} = \beta _{U}\mathrm {e}^{\tilde{z}_{2}-\varDelta -\sigma } + a(1-\mathrm {e}^{-\sigma })\) and this combined with \(\tilde{T}=\tilde{z}_{2}+\tau \) from (9) in the solution (44) computed at \(t=\tilde{T}\) yields
Hence the inequality \(x^{(\varDelta )}(\tilde{T})<0\) holds if and only if
Defining \(\delta _{4}\) as the constant such that \(x^{(\varDelta )}(\tilde{T})=0\) for \(\varDelta =\delta _{4}\) leads to (15).
Then \(x^{(\varDelta )}(\tilde{T})<0\) if, and only if, \(\varDelta <\delta _{4}\) while \(x^{(\varDelta )}(\tilde{T})\geqslant 0\) if, and only if, \(\varDelta \geqslant \delta _{4}\). The interval \({ I_{FNFP} }=(\tilde{z}_{2},\tilde{T}-\sigma )\cap (-\infty ,\delta _{2})\) combined with the condition \(x^{(\varDelta )}(\tilde{T})\geqslant 0\) (see the examples in Fig. 2) gives the interval \({ I_{FNFP1} }\) given by (13), while \({ I_{FNFP} }\) combined with the condition \(x^{(\varDelta )}(\tilde{T})<0\) (see the examples in Fig. 3) yields the interval \({ I_{FNFP2} }\) given by (14). \(\square \)
Proof of Remark 3.2
FNFP3: for this subcase we have to assume \(x^{(\varDelta )}(z_{\varDelta ,3}+\tau )< 0\) (see the examples in Fig. 4).
From \(x^{(\varDelta )}(\varDelta )<0\) and \(x^{(\varDelta )}(\varDelta +\sigma )>0\) we obtain a zero \(z_{\varDelta ,3}\) of \(x^{(\varDelta )}\) in \((\varDelta ,\varDelta +\sigma )\) given by
From (44) and \((x^{(\varDelta )}(\varDelta +\sigma )+\beta _{U})>0\) it follows that \(x^{(\varDelta )}(t)\) is strictly decreasing for \(t\in [\varDelta +\sigma ,\tilde{T}]\). This together with \(x^{(\varDelta )}(\varDelta +\sigma )>0\) and \(x^{(\varDelta )}(\tilde{T})<0\) yields that there is a zero \(z_{\varDelta ,4}\) of \(x^{(\varDelta )}\) in \((\varDelta +\sigma ,\tilde{T})\) given by
For \(t\in (\tilde{T},z_{\varDelta ,3}+\tau )\) we have \(\tilde{z}_{2}<t-\tau <z_{\varDelta ,3}\). Hence \(x^{(\varDelta )}(t-\tau )<0\) and
From (48) and \((x^{(\varDelta )}(\tilde{T})-\beta _{L})<0\) we obtain that \(x^{(\varDelta )}(t)\) is strictly decreasing for \(t\in (\tilde{T},z_{\varDelta ,3}+\tau )\). At \(t=(z_{\varDelta ,3}+\tau )\) Eq. (48) gives
which is negative by assumption.
The condition \(x^{(\varDelta )}(z_{\varDelta ,3}+\tau )< 0\) together with Eqs. (49), (46), (45) and \(a>\beta _{U}\) yield
Thus the interval \({ I_{FNFP2} }\), given by (14), together with the extra condition \(x^{(\varDelta )}(z_{\varDelta ,3}+\tau )<0\), written as (50), yield the interval \({ I_{FNFP3} }\) given by (16).
FNFP4: in this subcase we also have to assume \(x^{(\varDelta )}(z_{\varDelta ,4}+\tau )<0\) (see the examples in Fig. 3).
For \(t\in [z_{\varDelta ,3}+\tau ,z_{\varDelta ,4}+\tau ]\) we have \(z_{\varDelta ,3}\leqslant t-\tau \leqslant z_{\varDelta ,4}\). Hence \(x^{(\varDelta )}(t-\tau )\geqslant 0\) and
From the condition \(x^{(\varDelta )}(z_{\varDelta ,3}+\tau )< 0\) we have \(x^{(\varDelta )}(z_{\varDelta ,3}+\tau )+\beta _{U}>0\), so \(x^{(\varDelta )}\) is strictly decreasing on \([z_{\varDelta ,3}+\tau ,z_{\varDelta ,4}+\tau ]\), and
The condition \(x^{(\varDelta )}(z_{\varDelta ,4}+\tau )<0\) along with Eqs. (51), (47) and (49) yield
From (50) we infer that the condition \(x^{(\varDelta )}(z_{\varDelta ,3}+\tau )\geqslant 0\) implies \(\varDelta \geqslant \hat{\delta }_{4}\). This condition together with the interval \({ I_{FNFP2} }\), given by (14), plus the extra condition \(x^{(\varDelta )}(z_{\varDelta ,4}+\tau )<0\), written as (52), yield the interval \({ I_{FNFP4} }\) given by (17). \(\square \)
Proof of Remark 3.3
For each case (11) we compute the resetting time \(F(\varDelta )\) as follows:
RNRN:\(F(\varDelta )=\sigma \), since \(x^{(\varDelta )}(t)=\tilde{x}(t+(\tilde{z}_{1}-z_{\varDelta ,1}))\) for all \(t\geqslant \varDelta +\sigma \) and \(x^{(\varDelta )}(z_{\varDelta ,1}+\tau )=\tilde{x}(\tilde{z}_{1}+\tau )=\bar{x}\), where \(z_{\varDelta ,1}=\tilde{z}_{1}+T(\varDelta )-\tilde{T}\) and \(T(\varDelta )\) is as in Mackey et al. (2017, Proposition 5.1);
RNRP:\(F(\varDelta )=\tilde{z}_{1}+\tau +(z_{\varDelta ,2}-\tilde{z}_{2})-\varDelta \), since \(x^{(\varDelta )}(t)=\tilde{x}(t-(z_{\varDelta ,2}-\tilde{z}_{2}))\) for all \(t\geqslant \tilde{z}_{1}+\tau +(z_{\varDelta ,2}-\tilde{z}_{2})\) and \(x^{(\varDelta )}(z_{\varDelta ,2}+\tau )=\tilde{x}(\tilde{z}_{2}+\tau )={\underline{x}}\), where \(z_{\varDelta ,2}=\tilde{z}_{2}+T(\varDelta )-\tilde{T}\) and \(T(\varDelta )\) is as in Mackey et al. (2017, Proposition 5.2);
RPRP:\(F(\varDelta )=\tilde{z}_{1}+\tau +(z_{\varDelta ,2}-\tilde{z}_{2})-\varDelta \), since \(x^{(\varDelta )}(t)=\tilde{x}(t-(z_{\varDelta ,2}-\tilde{z}_{2}))\) for all \(t\geqslant \tilde{z}_{1}+\tau +(z_{\varDelta ,2}-\tilde{z}_{2})\) and \(x^{(\varDelta )}(z_{\varDelta ,2}+\tau )=\tilde{x}(\tilde{z}_{2}+\tau )={\underline{x}}\), where \(z_{\varDelta ,2}=\tilde{z}_{2}+T(\varDelta )-\tilde{T}\) and \(T(\varDelta )\) is as in Mackey et al. (2017, Proposition 5.3);
RPFP:\(F(\varDelta )=\sigma \), since \(x^{(\varDelta )}(t)=\tilde{x}(t-(z_{\varDelta ,2}-\tilde{z}_{2}))\) for all \(t\geqslant \varDelta +\sigma \) and \(x^{(\varDelta )}(z_{\varDelta ,2}+\tau )=\tilde{x}(\tilde{z}_{2}+\tau )={\underline{x}}\), where \(z_{\varDelta ,2}=\tilde{z}_{2}+T(\varDelta )-\tilde{T}\) and \(T(\varDelta )\) is as in Mackey et al. (2017, Proposition 5.4);
RPFN:\(F(\varDelta )=z_{\varDelta ,2}+\tau -\varDelta \), since \(x^{(\varDelta )}(t)=\tilde{x}(t-(z_{\varDelta ,3}-\tilde{z}_{3}))\) for all \(t\geqslant z_{\varDelta ,2}+\tau \) and \(x^{(\varDelta )}(z_{\varDelta ,3}+\tau )=\tilde{x}(\tilde{z}_{3}+\tau )=\bar{x}\), where \(z_{\varDelta ,3}=\tilde{z}_{3}+T(\varDelta )-\tilde{T}\) and \(T(\varDelta )\) is as in Mackey et al. (2017, Proposition 5.5);
FPFP:\(F(\varDelta )=\sigma \), since \(x^{(\varDelta )}(t)=\tilde{x}(t-(z_{\varDelta ,2}-\tilde{z}_{2}))\) for all \(t\geqslant \varDelta +\sigma \) and \(x^{(\varDelta )}(z_{\varDelta ,2}+\tau )=\tilde{x}(\tilde{z}_{2}+\tau )={\underline{x}}\), where \(z_{\varDelta ,2}=\tilde{z}_{2}+T(\varDelta )-\tilde{T}\) and \(T(\varDelta )\) is as in Mackey et al. (2017, Proposition 5.6);
FPFN:\(F(\varDelta )=z_{\varDelta ,2}+\tau -\varDelta \), since \(x^{(\varDelta )}(t)=\tilde{x}(t-(z_{\varDelta ,3}-\tilde{z}_{3}))\) for all \(t\geqslant z_{\varDelta ,2}+\tau \) and \(x^{(\varDelta )}(z_{\varDelta ,3}+\tau )=\tilde{x}(\tilde{z}_{3}+\tau )=\bar{x}\), where \(z_{\varDelta ,2}=\tilde{z}_{2}+T(\varDelta )-\tilde{T}\) and \(z_{\varDelta ,3}=\tilde{z}_{3}+T(\varDelta )-\tilde{T}\) with \(T(\varDelta )\) and \(z_{\varDelta ,2}\) given by Mackey et al. (2017, Proposition 5.7 and its proof);
FNFP1:\(F(\varDelta )=z_{\varDelta ,3}+\tau -\varDelta \), since \(x^{(\varDelta )}(t)=\tilde{x}(t-(z_{\varDelta ,4}-\tilde{z}_{4}))\) for all \(t\geqslant z_{\varDelta ,3}+\tau \) and \(x^{(\varDelta )}(z_{\varDelta ,4}+\tau )=\tilde{x}(\tilde{z}_{4}+\tau )={\underline{x}}\), where \(z_{\varDelta ,3}\) and \(z_{\varDelta ,4}=\tilde{z}_{4}\) are computed as follows. From \(x^{(\varDelta )}(\varDelta )<0<x^{(\varDelta )}(\varDelta +\sigma )\) we obtain a zero \(z_{\varDelta ,3}\) of \(x^{(\varDelta )}\) in \((\varDelta ,\varDelta +\sigma )\) given by
where \(x^{(\varDelta )}(\varDelta )=-\beta _{U}+\beta _{U}\mathrm {e}^{-(\tilde{z}_{2}-\varDelta )}\). For \(\tilde{T}<t<z_{\varDelta ,3}+\tau \) we have \(\tilde{z}_{2}<t-\tau <z_{\varDelta ,3}\). Hence \(x^{(\varDelta )}(t-\tau )<0\), and
Since \(\tilde{z}_{2}<z_{\varDelta ,3}\) and from the proof of Remark 3.1 the inequality \(\varDelta \geqslant \delta _{4}\) implies \(x^{(\varDelta )}(\tilde{T})\geqslant 0\), we obtain
Notice that \(x^{(\varDelta )}(t)>0\) on \((z_{\varDelta ,3},z_{\varDelta ,3}+\tau ]\). Using this and \(z_{\varDelta ,3}<\varDelta +\sigma<\tilde{z}_{2}+\tau <z_{\varDelta ,3}+\tau \) we obtain that \(x^{(\varDelta )}(t)\) is strictly decreasing on \([z_{\varDelta ,3}+\tau ,\infty )\) as long as \(x^{(\varDelta )}(t-\tau )\geqslant 0\). It follows that there is a first zero \(z_{\varDelta ,4}\) of \(x^{(\varDelta )}(t)\) in \([z_{\varDelta ,3}+\tau ,\infty )\) given by
with \(x^{(\varDelta )}(z_{\varDelta ,3}+\tau )\) given by (53), where \(x^{(\varDelta )}(\tilde{T})\) is given by (45);
FNFN:\(F(\varDelta )=\tilde{z}_{2}+\tau -\varDelta \), since \(x^{(\varDelta )}(t)=\tilde{x}(t+(\tilde{z}_{3}-z_{\varDelta ,3}))\) for all \(t\geqslant \tilde{z}_{2}+\tau \) and \(x^{(\varDelta )}(z_{\varDelta ,3}+\tau )=\tilde{x}(\tilde{z}_{3}+\tau )=\bar{x}\), where \(z_{\varDelta ,3}=\tilde{z}_{3}+T(\varDelta )-\tilde{T}\) and \(T(\varDelta )\) is as in Mackey et al. (2017, Proposition 5.8);
FNRN:\(F(\varDelta )=\sigma \), since \(x^{(\varDelta )}(t)=\tilde{x}(t+(\tilde{z}_{3}-z_{\varDelta ,3}))\) for all \(t\geqslant \varDelta +\sigma \) and \(x^{(\varDelta )}(z_{\varDelta ,3}+\tau )=\tilde{x}(\tilde{z}_{3}+\tau )=\bar{x}\), where \(z_{\varDelta ,3}=\tilde{z}_{3}+T(\varDelta )-\tilde{T}\) and \(T(\varDelta )\) is as in Mackey et al. (2017, Proposition 5.10);
FNRP:\(F(\varDelta )=\tilde{z}_{3}+\tau -(\tilde{z}_{4}-z_{\varDelta ,4})-\varDelta \), since \(x^{(\varDelta )}(t)=\tilde{x}(t+(\tilde{z}_{4}-z_{\varDelta ,4}))\) for all \(t\geqslant \tilde{z}_{3}+\tau -(\tilde{z}_{4}-z_{\varDelta ,4})\) and \(x^{(\varDelta )}(z_{\varDelta ,4}+\tau )=\tilde{x}(\tilde{z}_{4}+\tau )={\underline{x}}\), where \(z_{\varDelta ,4}=\tilde{z}_{4}+T(\varDelta )-\tilde{T}\) and \(T(\varDelta )\) is as in Mackey et al. (2017, Proposition 5.10). \(\square \)
Proof of Remark 3.4
Define the constants \(\delta _{1}\) as in Mackey et al. (2017, Eq. (5.6))
For each case (11) we consider the corresponding \(\varDelta \) interval as computed in Mackey et al. (2017, Section 5) and listed in (11). Recalling that \(\sigma \in (0,\tau ]\), \(\tilde{T}=\tilde{z}_{2}+\tau \), \(t_{max}=\tilde{z}_{1}+\tau \), \(\tilde{z}_{n}=\tilde{T}+\tilde{z}_{n-2}\) for \(n\in \{2,3,4\ldots \}\), \(\tilde{z}_{j+1}>\tilde{z}_{j}+\tau \) for all \(j\in \mathbb {N}\), and from Mackey et al. (2017, Proposition 4.2) we see that \(J=j_\varDelta \in \{0,1,2\}\) implies \(z_{\varDelta ,J+1}>\varDelta \) and \(z_{\varDelta ,J+1}>z_{\varDelta ,J+1}+\tau \). Thus, we show that for each case (11) we have \(F(\varDelta )<T(\varDelta )\) as follows:
RNRN:\(\varDelta \in { I_{RNRN} }=[0,\delta _{1}]\): since \(\tilde{z}_{2}>\tilde{z}_{1}+\tau >\tilde{z}_{1}\),
RNRP and RPRP:\(\varDelta \in { I_{RNRP} }=[\max \{0,\delta _{1}\},\tilde{z}_{1})\) with \(\delta _{1}\) given by (54) and \(\varDelta \in { I_{RPRP} }=[\tilde{z}_{1},t_{max}-\sigma ]\): using that \(\tilde{z}_{2}>\tilde{z}_{1}+\tau >\tilde{z}_{1}\) we infer
RPFP:\(\varDelta \in { I_{RPFP} }=(t_{\max }-\sigma ,t_{\max }]\cap (-\infty ,\delta _{2}]\):
RPFN:\(\varDelta \in { I_{RPFN} }=(t_{\max }-\sigma ,t_{\max }]\,\cap \,(\delta _{2},\infty )\): from Mackey et al. (2017, Proposition 4.2) we have \(z_{\varDelta ,3}>z_{\varDelta ,2}+\tau \), and thereby
and using the lower bound of \(\varDelta \) in \({ I_{RPFN} }\) we conclude that
FPFP:\(\varDelta \in { I_{FPFP} }=[t_{\max },\tilde{z}_{2}]\,\cap \,(-\infty ,\delta _{2}]\):
FPFN:\(\varDelta \in { I_{FPFN} }=[t_{\max },\tilde{z}_{2}]\,\cap \,(\delta _{2},\infty )\): using \(z_{\varDelta ,3}>z_{\varDelta ,2}\) and using the lower bound of \(\varDelta \) in \({ I_{FPFN} }\) we have
FNFP1:\(\varDelta \in { I_{FNFP1} }=\varDelta \in (\tilde{z}_{2},\tilde{T}-\sigma )\,\cap \,(-\infty ,\delta _{2})\,\cap \,[\delta _{4},\infty )\): here we have \(x^{(\varDelta )}(z_{\varDelta ,4}+t)=\tilde{x}(\tilde{z}_{2}+t)\) for all \(t\geqslant 0\), which gives \(T(\varDelta )= z_{\varDelta ,4}-\tilde{z}_{2}\). Noting that \(z_{\varDelta ,4}>z_{\varDelta ,3}\) and using the lower bound of \(\varDelta \) in \({ I_{FNFP1} }\) we obtain
FNFP2: Recall that \({ I_{FNFP2} }=\emptyset \) since \(\delta _{4}<\tilde{z}_{2}\);
FNFN:\(\varDelta \in { I_{FNFN} }=(\tilde{z}_{2},\tilde{z}_{2}+\tau -\sigma )\,\cap \,[\delta _{2},\infty )\): since \(z_{\varDelta ,3}>z_{\varDelta ,2}+\tau =\tilde{z}_{2}+\tau \) we have
FNRN:\(\varDelta \in { I_{FNRN} }=[\tilde{T}-\sigma ,\tilde{T})\,\cap \,(-\infty ,\tilde{T}+\delta _{1})\): using the fact that \(z_{\varDelta ,3}>z_{\varDelta ,2}+\tau =\tilde{z}_{2}+\tau \) we obtain
FNRP:\(\varDelta \in { I_{FNRP} }=[\tilde{T}-\sigma ,\tilde{T})\,\cap \,[\tilde{T}+\delta _{1},\infty )\): using that \(\tilde{z}_{2}+\tau>\tilde{z}_{1}+2\tau >\tilde{z}_{1}+\tau +\sigma \) we have \(\tilde{z}_{1}+\tau -(\tilde{T}-\sigma )<0\), and thus
\(\square \)
Proof of Remark 3.5
Each set of parameters \((\tau ,\beta _{U},\beta _{L},\sigma ,a,\varDelta )\) defines a sequence of cases (11) along \(\varDelta \in [0,\tilde{T})\). Thus we show that for each case (11) we have \(F(\varDelta )\geqslant \sigma \):
RNRN, RPFP, FPFP, FNRN:\(F(\varDelta )=\sigma \);
RNRP: from Mackey et al. (2017, Proposition 5.2) we have \(x^{(\varDelta )}(z_{\varDelta ,1}+\tau )>\bar{x}\). Using this and \(\bar{x}+\beta _{U}=\beta _{U}\mathrm {e}^{{\tilde{z}}_{2}-t_{max}}\) in \(0=x^{(\varDelta )}(z_{\varDelta ,2})=-\beta _{U}+(x^{(\varDelta )}(z_{\varDelta ,1}+\tau )+\beta _{U})\mathrm {e}^{z_{\varDelta ,1}+\tau -z_{\varDelta ,2}}\) we obtain
So \((z_{\varDelta ,2}-\tilde{z}_{2})>(z_{\varDelta ,1}-\tilde{z}_{1})\) and \(z_{\varDelta ,1}>\varDelta \) lead to
RPRP: taking the upper bound of \(\varDelta \) in \({ I_{RPRP} }=[\tilde{z}_{1},t_{max}-\sigma ]\) and using that \(z_{\varDelta ,2}>\tilde{z}_{2}\), see Mackey et al. (2017, Proposition 5.3), we find
RPFN: taking the upper bound of \(\varDelta \) in \({ I_{RPFN} }=(t_{\max }-\sigma ,t_{\max }]\,\cap \,(\delta _{2},\infty )\) and using that \(z_{\varDelta ,2}\geqslant \tilde{z}_{2}\) Mackey et al. (2017, Proof of Proposition 5.5, Eq. (9.5)) we obtain
FPFN: from Mackey et al. (2017, Proof of Proposition 5.7) we have \(z_{\varDelta ,2}\geqslant \tilde{z}_{2}\), then
FNFP1:\(F(\varDelta )=z_{\varDelta ,3}+\tau -\varDelta>z_{\varDelta ,3}+\tau -(\tilde{z}_{2}+\tau -\sigma )=\sigma +z_{\varDelta ,3}-\tilde{z}_{2}>\sigma \);
FNFP2: Recall that \({ I_{FNFP2} }=\emptyset \) since \(\delta _{4}<\tilde{z}_{2}\);
FNFN:\(F(\varDelta )=\tilde{z}_{2}+\tau -\varDelta >\tilde{z}_{2}+\tau -(\tilde{z}_{2}+\tau -\sigma )=\sigma \);
FNRP: from Mackey et al. (2017, Proof of Proposition 5.10) we have \(x^{(\varDelta )}(z_{\varDelta ,3}+\tau )>\bar{x}\). Using this and \(\bar{x}+\beta _{U}=\beta _{U}\mathrm {e}^{\tilde{z}_{2}-t_{max}}\) in \(x^{(\varDelta )}(z_{\varDelta ,4})=0\) we have
So \((z_{\varDelta ,4}-\tilde{z}_{4})>(z_{\varDelta ,3}-\tilde{z}_{3})\) and \(z_{\varDelta ,3}>\varDelta \) lead to
\(\square \)
Proof of Remark 3.6
Recall that \(\delta _{2}\) is given by (12) and is defined for \(\beta _{U}>a(1-\mathrm {e}^{-\sigma })\). Thus the definition (12) also holds for \(a<\beta _{U}\). For the case RPFN of Mackey et al. (2017, Table 2) we have \(\varDelta \in { I_{RPFN} }=(t_{\max }-\sigma ,t_{\max }]\cap (\delta _{2},\infty )\) and \(\delta _{2}>t_{max}\), then it follows that \({ I_{RPFN} }=\emptyset \). \(\square \)
Proof of Remark 3.7
From Mackey et al. (2017, Corollary 4.2) it follows that the cycle length map \(T(\varDelta )\) is continuous for \(a<\beta _{U}\). The proof is divided into two subcases, \(\delta _{2}<t_{max}\) and \(\delta _{2}\geqslant t_{max}\) as follows.
The condition \(\delta _{2}<t_{max}\) implies that \(a<\beta _{U}\) and there exists a sequence of cases from (11) as is shown in Mackey et al. (2017, Table 1). Once the cycle length map is continuous we see from Mackey et al. (2017, Third row of Table 1) that \(T(\varDelta )\) is strictly increasing on \([0,\delta _{2}]\) and strictly decreasing on \([\delta _{2},\tilde{T}]\). Thus the maximum of \(T(\varDelta )\) occurs for the case RPFP with \(\varDelta =\delta _{2}\) and we have \(\bar{T}=T(\delta _{2})\) with the cycle length map given by Mackey et al. (2017, Proposition 5.4), i.e.
where \(\tilde{T}\) is defined by (9).
For \(\delta _{2}\geqslant t_{max}\) there exists a sequence of cases from (11) as is shown in Mackey et al. (2017, Table 2) and it follows from Remark 3.6 that \(a<\beta _{U}\) holds. For the case RPFN of Mackey et al. (2017, Table 2) we have \(\delta _{2}\geqslant t_{max}\) and from Remark 3.6 it follows that \({ I_{RPFN} }=\emptyset \). Once the cycle length map is continuous and \({ I_{RPFN} }=\emptyset \), we see from Mackey et al. (2017, Third row of Table 2) that again \(T(\varDelta )\) is strictly increasing on \([0,\delta _{2}]\) and strictly decreasing on \([\delta _{2},\tilde{T}]\). Thus the maximum of \(T(\varDelta )\) occurs for the case FPFP with \(\varDelta =\delta _{2}\) and we have \(\bar{T}=T(\delta _{2})\) with the cycle length map given by Mackey et al. (2017, Proposition 5.6), which is equal to (55). \(\square \)
Proof of Remark 3.8
For \(\varDelta \leqslant t \leqslant \varDelta +\sigma \), Eq. (41) together with \(x^{(\varDelta )}(\varDelta ) = -\beta _{U} + \beta _{U}\mathrm {e}^{\tilde{z}_{2}-\varDelta }\) yields
The conditions \(a>\beta _{U}\) and \(\varDelta >\tilde{z}_{2}\) (see Remark 3.1) combined yield \(a>\beta _{U}\mathrm {e}^{\tilde{z}_{2}-\varDelta }\). Hence \(x^{(\varDelta )}(t)\) is strictly increasing on \([\varDelta ,\varDelta +\sigma ]\). From \(x^{(\varDelta )}(\varDelta )<0\) and \(x^{(\varDelta )}(\varDelta +\sigma )>0\) we obtain a zero \(z_{\varDelta ,3}\) of \(x^{(\varDelta )}\) in \((\varDelta ,\varDelta +\sigma )\) given by (46).
For \(\varDelta +\sigma \leqslant t \leqslant \tilde{T}\), Eq. (44) together with \((x^{(\varDelta )}(\varDelta +\sigma )+\beta _{U})>0\) shows that \(x^{(\varDelta )}(t)\) is strictly decreasing on \([\varDelta +\sigma ,\tilde{T}]\).
For \(\tilde{T}<t<z_{\varDelta ,3}+\tau \) we have \(\tilde{z}_{2}<t-\tau <z_{\varDelta ,3}\). Hence \(x^{(\varDelta )}(t-\tau )<0\), and
Thus \(x^{(\varDelta )}\) is strictly increasing on \([\tilde{T},z_{\varDelta ,3}+\tau ]\) since \(x^{(\varDelta )}(\tilde{T})<0\). For \(t=z_{\varDelta ,3}+\tau \)
A rapidly oscillating periodic solution occurs if \(x^{(\varDelta )}(\varDelta )=x^{(\varDelta )}(\tilde{T})\), \(x^{(\varDelta )}(\varDelta +\sigma )=x^{(\varDelta )}(z_{\varDelta ,3}+\tau )\) and if the solution \(x^{(\varDelta )}\) for \(\varDelta \leqslant t\leqslant \varDelta +\sigma \) is equal to the solution \(x^{(\varDelta )}\) for \(\tilde{T}\leqslant t\leqslant z_{\varDelta ,3}+\tau \), i.e, \(a=\beta _{L}+\beta _{U}\). Then, the necessary conditions for the existence of a rapid oscillation are
Combining Eq. (58) with (57) we get
Using the first and second relation in the third line of Eq. (59) gives
and combining this with (42) yields
From \(a>\beta _{U}\) and \(x^{(\varDelta )}(\varDelta ) = -\beta _{U} + \beta _{U}\mathrm {e}^{\tilde{z}_{2}-\varDelta }\) it follows that \((x^{(\varDelta )}(\varDelta ) + \beta _{U} - a)<0\). Hence the conditions (58) are reduced to
The relation (60) combined with (46) yields
where \(\tilde{z}_{2}\) is given by (9). Substituting \(\varDelta =\delta _{\infty }\) in (61) gives the constant defined \(\delta _{\infty }\) in (19) and
So the conditions (58) yield \(\varDelta =\delta _{\infty }\) with \(\tilde{z}_{2}<\delta _{\infty }<\tilde{z}_{2}+\sigma \).
The period of the unstable periodic solution is given by \(\tilde{T}^{(\infty )}=z_{\varDelta ,3}+\tau -(\varDelta +\sigma )\) (see the example from Fig. 5). Computing \(z_{\varDelta ,3}\) from (60) and using \(\varDelta =\delta _{\infty }\) gives \(\tilde{T}^{(\infty )}=\tilde{T}-\delta _{\infty }\), and this together with \(\tilde{z}_{2}<\delta _{\infty }<\tilde{z}_{2}+\sigma \) yields \(\tau -\sigma<\tilde{T}^{(\infty )}<\tau \).
Recalling that \(x^{(\varDelta )}(t)\) is strictly increasing on \([\varDelta ,\varDelta +\sigma ]\), strictly decreasing on \([\varDelta +\sigma ,\tilde{T}]\) and strictly increasing on \([\tilde{T},z_{\varDelta ,3}+\tau ]\), we infer that for \(\varDelta =\delta _{\infty }\) the minimum and maximum of the rapid limit cycle are respectively given by \({\underline{x}}^{(\varDelta )}=x^{(\varDelta )}(\delta _{\infty })\) and \(\bar{x}^{(\varDelta )}=x^{(\varDelta )}(\delta _{\infty }+\sigma )\). From \(x^{(\varDelta )}(\varDelta ) = -\beta _{U} + \beta _{U}\mathrm {e}^{\tilde{z}_{2}-\varDelta }\) with \(\varDelta =\delta _{\infty }\) and \(0<\delta _{\infty }-\tilde{z}_{2}<\sigma <\tau \) it follows that
From (56) with \(t=\delta _{\infty }+\sigma \) and \(\varDelta =\delta _{\infty }\) together with \(a=\beta _{L}+\beta _{U}\) and \(0<\delta _{\infty }-\tilde{z}_{2}<\sigma <\tau \) it follows that
\(\square \)
Proof of Proposition 4.1
Using the fact that \(x^{(p)}(\varDelta _{0})=x^{(p)}(z_{2})=0\), for \(t\in [\varDelta _{0},\varDelta _{0}+\sigma ]\) we have
The condition \(a\geqslant a_{1}\) implies \(a>\beta _{U}\), so \(x^{(p)}(t)\) is increasing and
thus \(x^{(p)}(\varDelta _{0}+\sigma ) = (a-\beta _{U})(1-\mathrm {e}^{-\sigma })>0\).
For \(t\in [\varDelta _{0}+\sigma ,\varDelta _{1}]\) we have
so \(x^{(p)}(t)\) is decreasing, since \((x^{(p)}(\varDelta _{0}+\sigma ) + \beta _{U}) = \beta _{U}\mathrm {e}^{-\sigma }+a(1-\mathrm {e}^{-\sigma })>0\), and
The condition \(a\geqslant a_{1}\) implies \(x^{(p)}(\varDelta _{1}) \geqslant 0\).
For \(t\in [\varDelta _{1},\varDelta _{1}+\sigma ]\) it follows that
so \(x^{(p)}(t)\) is increasing, \((x^{(p)}(\varDelta _{1})+\beta _{U}-a) = (\beta _{U}-a)\mathrm {e}^{-T_{p}}+a(\mathrm {e}^{-\alpha }-1)<0\), and
Since \(x^{(p)}(t)\) is increasing for \(t\in [\varDelta _{1},\varDelta _{1}+\sigma ]\) and \(x^{(p)}(\varDelta _{1}) \geqslant 0\), then \(x^{(p)}(\varDelta _{1}+\sigma )>x^{(p)}(\varDelta _{1})\geqslant 0\).
For \(t\in [\varDelta _{1}+\sigma ,\varDelta _{2}]\) the solution is given by
so \(x^{(p)}(t)\) is decreasing, \((x^{(p)}(\varDelta _{1}+\sigma )+\beta _{U}) = \beta _{U}\mathrm {e}^{-T_{p}-\sigma }+a(1-\mathrm {e}^{-\sigma })(1+\mathrm {e}^{-T_{p}})>0\), and
Thus \(x^{(p)}(\varDelta _{2}) > \beta _{U}(\mathrm {e}^{-T_{p}}-1)+a(1-\mathrm {e}^{-\sigma })\mathrm {e}^{-\alpha } = x^{(p)}(\varDelta _{1})\geqslant 0\).
For \(t\in [\varDelta _{2},\varDelta _{2}+\sigma ]\) we have
so \(x^{(p)}(t)\) is increasing, \((x^{(p)}(\varDelta _{2})+\beta _{U}-a) = (\beta _{U}-a)\mathrm {e}^{-2T_{p}}+a(\mathrm {e}^{-\alpha }-1)+(\mathrm {e}^{-\alpha }-1)\mathrm {e}^{-2\alpha }<0\), and
Since \(x^{(p)}(t)\) is increasing for \(t\in [\varDelta _{2},\varDelta _{2}+\sigma ]\) and \(x^{(p)}(\varDelta _{2})\geqslant 0\), then \(x^{(p)}(\varDelta _{2}+\sigma )>x^{(p)}(\varDelta _{2})>0\).
For \(t\in [\varDelta _{2}+\sigma ,\varDelta _{3}]\) it follows that
so \(x^{(p)}(t)\) is decreasing, \((x^{(p)}(\varDelta _{2}+\sigma )+\beta _{U}) = \beta _{U}\mathrm {e}^{-2T_{p}-\sigma }+a(1-\mathrm {e}^{-\sigma })[1+\mathrm {e}^{-T_{p}}+\mathrm {e}^{-2T_{p}}]>0\), and
Thus \(x^{(p)}(\varDelta _{3})> \beta _{U}(\mathrm {e}^{-T_{p}}-1)+a(1-\mathrm {e}^{-\sigma })\mathrm {e}^{-\alpha } = x^{(p)}(\varDelta _{2})>0\).
For \(t\in [\varDelta _{3},\varDelta _{3}+\sigma ]\) the solution is given by
so \(x^{(p)}(t)\) is increasing, \((x^{(p)}(\varDelta _{3})+\beta _{U}-a) = (\beta _{U}-a)\mathrm {e}^{-3T_{p}}+a(\mathrm {e}^{-\alpha }-1)+(\mathrm {e}^{-\alpha }-1)\mathrm {e}^{-3\alpha }<0\), and
Since \(x^{(p)}(t)\) is increasing for \(t\in [\varDelta _{3},\varDelta _{3}+\sigma ]\) and \(x^{(p)}(\varDelta _{3})\geqslant 0\), then \(x^{(p)}(\varDelta _{3}+\sigma )>x^{(p)}(\varDelta _{3})>0\).
Generalizing this procedure, we see that \(x^{(p)}(\varDelta _{n})\) and \(x^{(p)}(\varDelta _{n}+\sigma )\) can be written according to Eqs. (25) and (26). The proof is completed inductively for (25) and (26) with \(n=1,2,3\) and \(\varDelta _{n}=nT_{p}+\varDelta _{0}\). Hence, we see that for \(t\in [\varDelta _{n},\varDelta _{n}+\sigma ]\) the solution \(x^{(p)}(t)\) is given by Eq. (23) and for \(t\in [\varDelta _{n}+\sigma ,\varDelta _{n+1}]\) the solution \(x^{(p)}(t)\) is given by Eq. (24). \(\square \)
Proof of Proposition 4.2
(i): Since \(a\geqslant a_{1}\) we can take the limit \(n\longrightarrow \infty \) in Eqs. (25) and (26). We know that \(\sum _{k=0}^{\infty }y^{k}=1/(1-y)\) for \(|y|<1\), then
and
(ii): Using that \(x(0)=\varphi (0)={\underline{x}}^{(p)}\) and \(\varDelta _{0}=0\), for \(t\in [0,\sigma ]\) the solution is given by
once that \(x(t-\tau )=\varphi (t)\geqslant 0\). So \(x^{(p)}(t)\) is increasing, since \(({\underline{x}}^{(p)}+\beta _{U}-a)=a\mathrm {e}^{-\alpha }(1-\mathrm {e}^{-\sigma })/(1-\mathrm {e}^{-\sigma -\alpha })-a<0\), and
Using that \(x^{(p)}(\sigma ) = \bar{x}^{(p)}\), for \(t\in [\sigma ,\varDelta _{1}]\) we have
so \(x^{(p)}(t)\) is decreasing, since \((\bar{x}^{(p)} + \beta _{U}) = a(1-\mathrm {e}^{-\sigma })/(1-\mathrm {e}^{-T_{p}})>0\), and
Furthermore, once \(x^{(p)}(\varDelta _{1}) = {\underline{x}}^{(p)}\), for \(t\in [\varDelta _{1},\varDelta _{1}+\sigma ]\) it follows that
and thus \(x^{(p)}(\varDelta _{1}+\sigma ) = \bar{x}^{(p)}\). For \(t\in [\varDelta _{1}+\sigma ,\varDelta _{2}]\) we have
Hence, repeating this process for \(t\in [\varDelta _{2},\varDelta _{2}+\sigma ]\), \([\varDelta _{2}+\sigma ,\varDelta _{2}]\), \([\varDelta _{3},\varDelta _{3}+\sigma ]\) and so forth we see that the solution \(x^{(p)}(t)\) is given by (27). The proof is completed by checking that the Principle of Mathematical Induction holds for (27) with \(n=1,2,3\) and \(\varDelta _{n}=nT_{p}+\varDelta _{0}\). \(\square \)
Proof of Proposition 4.3
Consider the maximum (29) and define an initial perturbation phase \(\varDelta _{l}\) such that \(x^{(p)}(\varDelta _{l})={\underline{x}}^{(p)}\). The solution for the initial pulse is given by (41) (Mackey et al. 2017, Section 5.3) with \(x^{(\varDelta )}(\varDelta )=-\beta _{U}+\beta _{U}\mathrm {e}^{{\tilde{z}}_{2}-\varDelta }\). Thus \(x^{(p)}(\varDelta _{l}) = -\beta _{U} + \beta _{U}\mathrm {e}^{{\tilde{z}}_{2}-\varDelta _{l}}={\underline{x}}^{(p)}\). This with \({\underline{x}}^{(p)}\geqslant 0\), since \(a\geqslant a_{1}\), gives
The proof is divided between the four cases shown in Fig. 9, where each \(\varDelta _{0}\) interval is given by: (a) \((\tilde{z}_{1},t_{max})\), (b) \([t_{max},\varDelta _{l})\), (c) \([\varDelta _{l},\tilde{z}_{2})\), (d) \([\tilde{z}_{2},\tilde{T}+\tilde{z}_{1}]\).
First, we prove case (b) by showing that for \(\varDelta _{0}\in [t_{max},\varDelta _{l})\) the points \(x^{(p)}(\varDelta _{n})\) converge to (28) and the points \(x^{(p)}(\varDelta _{n}+\sigma )\) converge to (29). For this case \(x^{(p)}(t_{max})\geqslant x^{(p)}(\varDelta _{0})>{\underline{x}}^{(p)}\geqslant 0\) and \(x^{(p)}(\varDelta _{0}-\tau )\geqslant 0\). Thus for \(t=\varDelta _{0}\) we have
For \(t\in [\varDelta _{0},\varDelta _{0}+\sigma ]\) the solution is given by \(x^{(p)}(t)=-\beta _{U}+a+(x^{(p)}(\varDelta _{0})+\beta _{U}-a)\mathrm {e}^{-(t-\varDelta _{0})}\), so
For \(t\in [\varDelta _{0}+\sigma ,\varDelta _{1}]\) we have \(x^{(p)}(t)=-\beta _{U}+(x^{(p)}(\varDelta _{0}+\sigma )+\beta _{U})\mathrm {e}^{-(t-\varDelta _{0}-\sigma )}\), then
Repeating the previous steps for \(t\in [\varDelta _{1},\varDelta _{1}+\sigma ]\), \(t\in [\varDelta _{1}+\sigma ,\varDelta _{2}]\) and so forth, we see that for \(t\in [\varDelta _{n},\varDelta _{n}+\sigma ]\) and \(n\in \mathbb {N}_{>0}\) the solution is given by \(x^{(p)}(t)=-\beta _{U}+a+(x^{(p)}(\varDelta _{n})+\beta _{U}-a)\mathrm {e}^{-(t-\varDelta _{n})}\), so
and for \(t\in [\varDelta _{n}+\sigma ,\varDelta _{n+1}]\) and \(n\in \mathbb {N}_{>0}\) we have \(x^{(p)}(t)=-\beta _{U}+(x^{(p)}(\varDelta _{n}+\sigma )+\beta _{U})\mathrm {e}^{-(t-\varDelta _{n}-\sigma )}\), then
From (63), (64) and \(T_{p}=\alpha +\sigma \) it follows that
Equation (65) is recursive and can be rewritten as
It is known that \(\sum _{k=0}^{\infty }y^{k}=1/(1-y)\) for \(|y|<1\), thus taking the limit \(n\longrightarrow \infty \) of (66) gives
and taking the limit \(n\longrightarrow \infty \) of (63) and using (67) yields
Then in the limit \(n\longrightarrow \infty \) we see that the solution of (21) converges to the limit cycle given by (27).
For case (c) the proof is the same as the case (b), but with \(\varDelta _{0}\in [\varDelta _{l},\tilde{z}_{2})\) and hence \(x^{(p)}(\varDelta _{l})\geqslant x^{(p)}(\varDelta _{0})>x^{(p)}(\tilde{z}_{2})\geqslant 0\).
To prove case (a) we first note that \(x^{(p)}(t-\tau )<0\) for \(t\in (\tilde{z}_{1},t_{max})\). Given an initial perturbation phase \(\varDelta _{0}\in (\tilde{z}_{1},t_{max})\), the solution on \((\tilde{z}_{1},t_{max})\) alternates between \(x^{(p)}(t)=\beta _{L}+a+(x^{(p)}(\varDelta _{n})-\beta _{L}-a)\mathrm {e}^{-(t-\varDelta _{n})}\) for \([\varDelta _{n},\varDelta _{n}+\sigma ]\) and \(x^{(p)}(t)=\beta _{L}+(x^{(p)}(\varDelta _{0}+\sigma )-\beta _{L})\mathrm {e}^{-(t-\varDelta _{0}-\sigma )}\) for \([\varDelta _{n}+\sigma ,\varDelta _{n}]\) with \(n=0,1,2,\ldots ,k\), where the k-th index is such that \(\varDelta _{k-1}< t_{max} \leqslant \varDelta _{k}\). Along the interval \((\tilde{z}_{1},t_{max})\) the solution \(x^{(p)}(t)\) might have a maximum point if it reaches the value \((\beta _{L}+a)\) in the intervals \([\varDelta _{n},\varDelta _{n}+\sigma ]\) or if it reaches the value \(\beta _{L}\) in the intervals \([\varDelta _{n}+\sigma ,\varDelta _{n}]\), otherwise, it will be strictly increasing for \(t\in (\tilde{z}_{1},t_{max})\), as is in the examples of Fig. 9a. For \(t\geqslant t_{max}\) the proof is the same as the case (b), but using \(x^{(p)}(t_{max})\) as initial point \(x^{(p)}(\varDelta _{0})\).
For case (d) we need to distinguish it between two subcases, the interval \(t\in [\varDelta _{0},\tilde{T}]\), for which \(x^{(p)}(t-\tau )\geqslant 0\), and the interval \(t\in (\tilde{T},z_{p,1}+\tau )\), where \(x^{(p)}(t-\tau )<0\) and \(z_{p,1}\) is the first zero of \(x^{(p)}(t)\) with \(t>\varDelta _{0}\). For the first interval the solution alternates between \(x^{(p)}(t)=-\beta _{U}+a+(x^{(p)}(\varDelta _{n})+\beta _{U}-a)\mathrm {e}^{-(t-\varDelta _{n})}\) for \([\varDelta _{n},\varDelta _{n}+\sigma ]\) and \(x^{(p)}(t)=-\beta _{U}+(x^{(p)}(\varDelta _{0}+\sigma )+\beta _{U})\mathrm {e}^{-(t-\varDelta _{0}-\sigma )}\) for \([\varDelta _{n}+\sigma ,\varDelta _{n}]\) with \(n=0,1,2,\ldots ,k\), where the k-th interval is such that \(\varDelta _{k-1}<(z_{p,1}+\tau )\leqslant \varDelta _{k}\). For the second interval, the solution alternates between \(x^{(p)}(t)=\beta _{L}+a+(x^{(p)}(\varDelta _{n})-\beta _{L}-a)\mathrm {e}^{-(t-\varDelta _{n})}\) for \([\varDelta _{n},\varDelta _{n}+\sigma ]\) and \(x^{(p)}(t)=\beta _{L}+(x^{(p)}(\varDelta _{0}+\sigma )-\beta _{L})\mathrm {e}^{-(t-\varDelta _{0}-\sigma )}\) for \([\varDelta _{n}+\sigma ,\varDelta _{n}]\) with \(n=0,1,2,\ldots ,k\), where the k-th index is such that \(\varDelta _{k-1}<(\tilde{T}+\tilde{z}_{1})\leqslant \varDelta _{k}\). Along the interval \([\varDelta _{0},\tilde{T}]\) the solution \(x^{(p)}(t)\) might oscillate if it reaches the value \((a-\beta _{U})\) in the intervals \([\varDelta _{n},\varDelta _{n}+\sigma ]\) or if it reaches the value \(-\beta _{U}\) in the intervals \([\varDelta _{n}+\sigma ,\varDelta _{n}]\), otherwise, it will be strictly increasing for \(t\in [\varDelta _{0},\tilde{T}]\), as is in the examples of Fig. 9d. For \(t>(z_{p,1}+\tau )\) the proof is the same as the case (a), but using \(x^{(p)}(z_{p,1}+\tau )\) as initial point \(x^{(p)}(\varDelta _{0})\). \(\square \)
Rights and permissions
About this article
Cite this article
De Souza, D.C., Mackey, M.C. Response of an oscillatory differential delay equation to a periodic stimulus. J. Math. Biol. 78, 1637–1679 (2019). https://doi.org/10.1007/s00285-018-1322-y
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00285-018-1322-y
Keywords
- Delay differential equation
- Periodic perturbation
- Delayed negative feedback
- Cycle length map
- Resetting time
- Blood cells
- Dynamical disease
- Cyclical neutropenia
- Cyclical thrombocytopenia