1: %%%%===============================================================================
2: %%%% Title: Pair Approximation Models for Disease Spread
3: %%%% to be published in The European Physical Journal B
4: %%%% Author: Jerome BENOIT, Ana NUNES and Margarida TELO da GAMA
5: %%%% Corresponding author: Jerome BENOIT
6: %%%% Email: benoit@cii.fc.ul.pt
7: %%%% Address:
8: %%%% Complexo Interdisciplinar da U. L.
9: %%%% Av. Prof. Gama Pinto, 2
10: %%%% P-1649-003 Lisboa
11: %%%% Portugal
12: %%%%
13: %%%% files: `pamds.{tex,rty,bib}' (LaTeX2e+Springer-SVJour+AmSLaTeX);
14: %%%% figures: `pamds-PhaseDiagram-UPA.mps',
15: %%%% `pamds-PhaseDiagram-CPA.mps',
16: %%%% `pamds-EndemicInfectivePlots.mps',
17: %%%% (MetaPost PostScript).
18: %%%%
19: %%%%===============================================================================
20: %%%%@ARTICLE{PAMDS,
21: %%%% author = {J\'er\^ome Benoit and Ana Nunes and Margarida {Telo da Gama}},
22: %%%% title = {Pair Approximation Models for Disease Spread},
23: %%%% journal = EPJB,
24: %%%% year = {2005},
25: %%%% eprint = {q-bio/0510005},
26: %%%%}
27: %%%%===============================================================================
28: %% `pamds.tex' SVJour file | LaTeX it, don't read it!
29: %% Pair Approximation Models for Disease Spread
30: %%
31: %% last major modification: 2005/09/11
32: %% last minor modification: 2006/02/06
33: %%
34: %%% checksum = ""
35: %%% ASCII table
36: %%% Upper-case A B C D E F G H I J K L M N O P Q R S T U V W X Y Z
37: %%% Lower-case a b c d e f g h i j k l m n o p q r s t u v w x y z
38: %%% Digits 0 1 2 3 4 5 6 7 8 9
39: %%% Exclamation ! Double quote " Hash (number) #
40: %%% Dollar $ Percent % Ampersand &
41: %%% Acute accent ' Left paren ( Right paren )
42: %%% Asterisk * Plus + Comma ,
43: %%% Minus - Point . Solidus /
44: %%% Colon : Semicolon ; Less than <
45: %%% Equals = Greater than > Question mark ?
46: %%% At @ Left bracket [ Backslash \
47: %%% Right bracket ] Circumflex ^ Underscore _
48: %%% Grave accent ` Left brace { Vertical bar |
49: %%% Right brace } Tilde ~
50: %%
51: \NeedsTeXFormat{LaTeX2e}[1997/12/01]
52: %%
53: \documentclass[%
54: epj
55: ]{svjour}[1999/09/14]
56: \usepackage{amsmath}
57: \usepackage{amssymb}
58: %\usepackage{eufrak}
59: \usepackage{mathrsfs}
60: %\usepackage{slashed}
61: \usepackage[sectionbib,numbers,sort&compress]{natbib}%\providecommand\newblock{\ }
62: %\usepackage{cite}
63: \usepackage{graphicx}
64: \usepackage{color}
65: \ifx\pdfoutput\undefined
66: \DeclareGraphicsExtensions{.mps,.ps,.eps}
67: %\DeclareGraphicsRule{*}{ps}{*}{}
68: \providecommand{\eprint}[2][]{\texttt{#2}}%
69: \else
70: %\usepackage[print=true,edit=false,copy=false,annotate=false,set]{pdfcrypt}[2001/08/09]
71: \DeclareGraphicsExtensions{.mps,.ps,.eps}
72: \DeclareGraphicsRule{*}{mps}{*}{}
73: \usepackage{hyperref}
74: \hypersetup{%
75: pdftitle={Pair Approximation Models for Disease Spread},
76: pdfauthor={%
77: J\'er\^ome Benoit,
78: Ana Nunes and
79: Margarida Telo da Gama},
80: pdfsubject={PACS: 02.50.-r, 87.23.Ge, 05.70.Ln},
81: pdfkeywords={complex systems; epidemic models; phase transitions},
82: pdfpagemode=None,%None,UseThumbs,UseOutlines,FullScreen
83: pdfstartpage=1,
84: pdfhighlight=/O,%/I/N/O/P
85: pdfview=FitH,
86: pdfstartview=FitH,
87: %% citecolor=green,
88: urlcolor=blue,
89: %% backref=false,
90: %% pagebackref=false,
91: }
92: \IfFileExists{thumbpdf.sty}{\usepackage{thumbpdf}}{}
93: \providecommand{\eprint}[2][]{\href{http://arxiv.org/abs/#2}{\texttt{#2}}}%
94: \fi
95: \IfFileExists{datetime.sty}{%
96: \usepackage{datetime}
97: \providecommand{\now}{\xxivtime\space\today}
98: }{%
99: \providecommand{\now}{\the\time,\space\today}
100: }
101: %%\usepackage{comment}
102:
103: \hyphenation{ame-nable}
104:
105: %% Definitions:
106: \input{\jobname.rty}
107:
108: \begin{document}
109: %%\preprint{\textcolor[rgb]{1,0,0}{\texttt{DRAFT}}}
110: %%\preprint{\textcolor[rgb]{1,0,0}{$\beta$ \texttt{version}}}
111: %%\preprint{$\beta$ \texttt{revision}}
112: %%\preprint{\eprint{q-bio/0510005}}
113: \title{Pair Approximation Models for Disease Spread}
114: %%\subtitle{Pair Approximation Models for Disease Spread}
115: \author{%
116: J\'er\^ome Benoit\thanks{\email{benoit@cii.fc.ul.pt}}
117: \and
118: Ana Nunes
119: \and
120: Margarida Telo da Gama
121: }
122: %%\mail{benoit@cii.fc.ul.pt}
123: \institute{%
124: Centro de F\'isica Te\'orica e
125: Computacional e Departamento de F\'\i sica,\\
126: Faculdade de Ci\^encias da Universidade de Lisboa,\\
127: Avenida Professor Gama Pinto 2,
128: P-1649-003 Lisboa,
129: Portugal%
130: }
131: %%\date{\now}
132: \date{\eprint{q-bio/0510005}}
133: %%\date{Received: DATE / Revised version: DATE}
134: \abstract{%
135: We consider a Susceptible-Infective-Recovered (\textsc{SIR}) model,
136: where the mechanism for the renewal of susceptibles is demographic,
137: on a ring with next nearest neighbour interactions,
138: and a family of correlated pair approximations (\textsc{CPA}),
139: parametrized by a measure of the relative contributions of loops and open
140: triplets of the sites involved in the infection process.
141: We have found that the phase diagram of the \textsc{CPA},
142: at fixed coordination number,
143: changes qualitatively as the relative weight of the loops increases,
144: from the phase diagram of the uncorrelated pair approximation
145: to phase diagrams typical of one-dimensional systems.
146: In addition,
147: we have performed computer simulations of the same model and shown that
148: while the \textsc{CPA} with a constant correlation parameter
149: cannot describe the global behaviour of the model,
150: a reasonable description of the endemic equilibria
151: as well as of the phase diagram may be obtained
152: by allowing the parameter to depend on the demographic rate.
153: %%
154: %%%% suggested PACS numbers (1999):
155: %%%% xx.xx.xx XXX
156: \PACS{
157: {02.50.-r}{Probability theory, stochastic processes, and statistics}
158: \and
159: {87.23.Ge}{Dynamics of social systems}
160: \and
161: {05.70.Ln}{Nonequilibrium and irreversible thermodynamics}
162: }
163: %%%% suggested keywords
164: %% \keywords{complex systems -- epidemic models -- phase transitions}
165: }
166: \maketitle
167:
168: \section{Introduction}\label{sec/introduction}
169: Stochastic Susceptible-Infective-Recovered (\textsc{SIR}) epidemic models
170: on lattices and networks can be mapped on to percolation problems
171: and are well understood \cite{Grassberger1982,MooreNewman2000May,MooreNewman2000Nov}.
172: To describe disease spread and persistence in a community,
173: the model must be extended to include a mechanism for renewal of susceptibles,
174: either births or immunity waning.
175:
176: Models with immunity waning,
177: Susceptible-Infective-Recovered-Susceptible (\textsc{SIRS}),
178: are based on the following transitions:
179: \begin{equation}
180: \label{ESWN/SIRS/scheme}
181: {S}{\xrightarrow[I_{\eswnNeighbourIndex}]{\mspace{15.0mu}\eswnInfectionRate\mspace{15.0mu}}}%
182: {I}{\xrightarrow{\mspace{15.0mu}\eswnRecoveryRate\mspace{15.0mu}}}%
183: {R}{\xrightarrow{\mspace{15.0mu}\eswnDemographicRate\mspace{15.0mu}}}%
184: {S},
185: \end{equation}
186: meaning that
187: any susceptible individual $S$ can be infected
188: by an infected neighbour $I_{\eswnNeighbourIndex}$
189: at the infection rate $\eswnInfectionRate$,
190: any infected individual $I$ becomes recovered $R$
191: at the recovery rate $\eswnRecoveryRate$,
192: and
193: any recovered individual $R$ becomes susceptible $S$
194: at the immunity loss rate $\eswnDemographicRate$.
195: Following customary habits,
196: we shall choose time units
197: for which \mbox{$\eswnRecoveryRate\!=\!1$}.
198:
199: The \textsc{SIRS} model interpolates between two well known models,
200: the contact process
201: (also known as Susceptible-Infective-Susceptible or \textsc{SIS})
202: and the \textsc{SIR} model,
203: in the limits $\eswnDemographicRate\rightarrow\infty$ and $\eswnDemographicRate\rightarrow{0}$, respectively,
204: and much is known about its behaviour on regular lattices,
205: both from the point of view of rigorous results
206: \cite{DurrettNeuhauser1991,AndjelSchinazi1996,vdBerg1998}
207: and of assessing the performance of mean field
208: and pair approximations against stochastic simulations \cite{JooLebowitz2004}.
209:
210: In particular, it is known \cite{vdBerg1998} that
211: on hypercubic lattices of arbitrary dimension
212: the phase diagram of \eqref{ESWN/SIRS/scheme} has two critical values,
213: \mbox{$\eswnCriticalInfectionRate(\infty)<\eswnCriticalInfectionRate(0)$},
214: which are the critical rates of the two limit problems
215: that is the contact process and \textsc{SIR}, respectively.
216: For \mbox{$\eswnInfectionRate<\eswnCriticalInfectionRate(\infty)$}
217: there is disease extinction for every $\eswnDemographicRate$,
218: while for $\eswnCriticalInfectionRate(0)<\eswnInfectionRate$
219: there is disease persistence for every $\eswnDemographicRate$.
220: For $\eswnCriticalInfectionRate(\infty)<\eswnInfectionRate<\eswnCriticalInfectionRate(0)$
221: disease persistence occurs only for $\eswnDemographicRate$ above a certain threshold.
222: The region of disease persistence for every $\eswnDemographicRate$ is `missing'
223: in dimension \mbox{$\eswnDimension\!=\!1$},
224: because in this case $\eswnCriticalInfectionRate(0)$ is infinite.
225:
226: In \cite{JooLebowitz2004} the uncorrelated pair approximation
227: (\textsc{UPA}, see Section~\ref{sec/MFAUPA})
228: was applied to the \textsc{SIRS} model \eqref{ESWN/SIRS/scheme} on linear and square lattices,
229: and the phase diagrams computed from the corresponding equations of evolution were compared
230: with the mean field phase diagram and with the results of simulations.
231: It was shown that,
232: by contrast with the mean field approximation,
233: the \textsc{UPA} phase diagram agrees qualitatively
234: with the simulations and the exact results both
235: in \mbox{$\eswnDimension\!=\!1$} and in \mbox{$\eswnDimension\!=\!2$}.
236:
237: Since the \textsc{UPA} does not take into account the lattice dimensionality explicitly,
238: it predicts identical phase diagrams on lattices with the
239: same coordination number $\eswnDegree$,
240: namely on linear (\mbox{$\eswnDimension\!=\!1$})
241: and square (\mbox{$\eswnDimension\!=\!2$}) lattices,
242: when next nearest neighbours (\mbox{$\eswnDegree\!=\!4$}) are considered.
243: However,
244: in one dimension the critical infection rate \mbox{$\eswnCriticalInfectionRate(0)=\infty$},
245: and the critical line has an asymptote at $\eswnDemographicRate=0$, while
246: in two dimensions the critical line crosses the $\eswnDemographicRate=0$ axis at a
247: finite value of $\eswnCriticalInfectionRate$,
248: which is the result of the \textsc{UPA} for \mbox{$\eswnDegree\!=\!4$}.
249:
250: The question is then
251: whether generalized pair approximations can account
252: for the dependence on dimensionality and,
253: in particular,
254: whether they can describe phase diagrams
255: with different qualitative behaviours at fixed coordination numbers.
256:
257: We have addressed this question, and more generally the problem of
258: constructing suitable pair approximations (Section~\ref{sec/MFAUPA}),
259: in the context of a modification of model \eqref{ESWN/SIRS/scheme},
260: where the mechanism of renewal of susceptibles is demography,
261: rather than immunity waning.
262: This is the natural scenario in the epidemiology of diseases that
263: confer permanent immunity,
264: such as childhood infectious diseases \cite{AndersonMay,MurrayII}.
265: For this model infection obeys the same rules as in \eqref{ESWN/SIRS/scheme},
266: immunity is permanent and all individuals,
267: whatever their state,
268: are submitted to birth and death events at a rate $\eswnDemographicRate$.
269: The stochastic process,
270: which describes the dynamics of this system,
271: is governed by the transitions
272: \begin{subequations}
273: \label{ESWN/scheme}
274: \begin{gather}
275: \label{ESWN/scheme/disease}
276: {S}{\xrightarrow[I_{\eswnNeighbourIndex}]{\mspace{15.0mu}
277: \eswnInfectionRate\mspace{15.0mu}}}%
278: {I}{\xrightarrow{\mspace{15.0mu}\eswnRecoveryRate\mspace{15.0mu}}}{R}%
279: ,
280: \\
281: %%%%\intertext{while}%
282: \label{ESWN/scheme/demographic}
283: {\{S,I,R\}}{\xrightarrow{\mspace{15.0mu}\eswnDemographicRate\mspace{15.0mu}}}{S}
284: .
285: \end{gather}
286: \end{subequations}
287: In the limit $\eswnDemographicRate=0$,
288: both models \eqref{ESWN/SIRS/scheme} and \eqref{ESWN/scheme}
289: coincide with \textsc{SIR} model.
290: In the opposite limit the dynamics of the two models are drastically different.
291: While,
292: in the limit $\eswnDemographicRate=\infty$,
293: \textsc{SIRS} coincides with the contact model \cite{JooLebowitz2004},
294: in the same limit the dynamics of \eqref{ESWN/scheme} is trivial:
295: it is driven by demography,
296: that keeps the entire population susceptible for any $\eswnCriticalInfectionRate$,
297: and thus $\eswnCriticalInfectionRate(\infty)=\infty$.
298: We are interested in the regime,
299: where $\eswnDemographicRate$ is smaller than the recovery rate,
300: which is meaningful for the study of acute disease spread.
301: Although in this regime the dynamics is dominated by the infection and recovery processes
302: which are identical in both models,
303: the behaviour of \eqref{ESWN/scheme} appears to be different,
304: in a subtle way,
305: from that of \eqref{ESWN/SIRS/scheme} (Section~\ref{sec/MFAUPA}).
306:
307: We have considered the demographic \textsc{SIR} model \eqref{ESWN/scheme}
308: on a linear lattice with periodic boundary conditions (ring)
309: and next nearest neighbour interactions, \mbox{$\eswnDegree\!=\!4$}.
310: We constructed a family of correlated pair approximations (\textsc{CPA}),
311: parametrized by $\eswnClosedFormParameter$,
312: a measure of the relative contributions of loops and open triplets of
313: connected sites involved in the disease spread (Section~\ref{sec/CPA}).
314: For $\eswnClosedFormParameter=0$ the approximation reduces to the standard \textsc{UPA}
315: (Section~\ref{sec/MFAUPA}).
316: The phase diagrams of the \textsc{CPA} show that
317: as $\eswnClosedFormParameter$ increases from $0$ to $\eswnRemarkableClosedFormParameter$
318: (see Section~\ref{sec/CPA})
319: the \textsc{CPA} interpolates between the \mbox{$\eswnDegree\!=\!4$} \textsc{UPA} critical behaviour
320: and the typical one-dimensional phase behaviour,
321: with $\eswnCriticalInfectionRate(0)=\infty$.
322: Finally,
323: we have simulated the demographic \textsc{SIR} model \eqref{ESWN/scheme}
324: on a ring, with \mbox{$\eswnDegree\!=\!4$}.
325: The results of the simulations indicate that
326: while the \textsc{CPA} with a constant value of $\eswnClosedFormParameter$ cannot describe
327: the global phase diagram of \eqref{ESWN/scheme},
328: a reasonable description of endemic equilibria
329: as well as of the phase diagram is obtained
330: when $\eswnClosedFormParameter$ is allowed to depend on the demographic rate $\eswnDemographicRate$
331: (Section~\ref{sec/CPA}).
332: This illustrates that in addition to describe
333: the dimensional crossover for lattices with coordination number \mbox{$\eswnDegree\!=\!4$},
334: the \textsc{CPA} can be made semi-quantitative providing an alternative to
335: the stochastic simulations of individual based models.
336: We conclude in Section~\ref{sec/discussion} with a brief discussion of the results.
337:
338: \begin{figure}[t]
339: \begin{center}
340: \includegraphics[width=1.0\linewidth]{pamds-EndemicInfectivePlots}%
341: \end{center}
342: \caption{%% see `pamds/meta/pamds-EndemicInfectivePlot-???_???.mp.data.mbj'
343: Endemic infective probability versus infection rate
344: at (a) high demographic rate, and at (b) huge demographic rate:
345: the endemic infective probability is plotted
346: from simulations (open circles),
347: the \textsc{MFA} (long dashed lines),
348: the \textsc{UPA} (dashed dotted lines)
349: and the correlated model with best-fit closed form parameters (solid lines).
350: The fitting procedure is based on perpendicular offsets
351: and on the assumption that the closed form parameter $\eswnClosedFormParameter$
352: depends only on the demographic rate $\eswnDemographicRate$.
353: Closed form parameters $\eswnClosedFormParameter$
354: for (a) and (b) respectively:
355: $0.50$,
356: $0.70$.
357: }%
358: \label{fig/X/endemic}
359: \end{figure}
360:
361: \section{Mean Field and Uncorrelated Pair Approximations}\label{sec/MFAUPA}
362: In this section we consider the time evolution of the demographic \textsc{SIR}
363: model on regular lattices and review the mean-field and (standard) uncorrelated pair approximations,
364: setting the notation and the stage for the development of the more sophisticated correlated pair approximations.
365:
366: In the demographic \textsc{SIR} model on networks,
367: sites represent individuals and bonds social links.
368: The dynamics is governed by the stochastic process \eqref{ESWN/scheme}.
369: Denoting by $\prob{A}$ the probability for an individual to be in state $A$ (at time $t$),
370: $\prob{AB}$ the probability for a lattice bond to connect an individual in state $A$ to an individual in state $B$,
371: the time evolution of the singleton probabilities $\prob{A}$
372: %%The time evolution of the singleton probabilities $\prob{A}$,
373: %%required to satisfy the normalization condition
374: %%\begin{equation}
375: %%\label{ESWN/SOCL/1}
376: %% \prob{S}+\prob{I}+\prob{R}=1
377: %% ,
378: %%\end{equation}
379: can be described by the set of first order differential equations
380: \cite{AndersonMay,MurrayII}:
381: \begin{subequations}
382: \label{ESWN/SDE/1}
383: \begin{align}
384: \label{ESWN/SDE/1/S}
385: \difft{\prob{S}}&=%
386: +\eswnDemographicRate\:%
387: \bigl[%
388: \prob{I}%
389: +%
390: \prob{R}%
391: \bigr]
392: -\eswnInfectionRate\sum_{\eswnNeighbourIndex}{\prob{{S}{I_{\eswnNeighbourIndex}}}}
393: ,
394: \\
395: \label{ESWN/SDE/1/I}
396: \difft{\prob{I}}&=%
397: +\eswnInfectionRate\sum_{\eswnNeighbourIndex}{\prob{{S}{I_{\eswnNeighbourIndex}}}}
398: -(\eswnDemographicRate+\eswnRecoveryRate)\:%
399: \prob{I}
400: ,
401: \\
402: \label{ESWN/SDE/1/R}
403: \difft{\prob{R}}&=%
404: +\eswnRecoveryRate\:%
405: \prob{I}
406: -\eswnDemographicRate\:%
407: \prob{R}
408: ,
409: \end{align}
410: \end{subequations}
411: where the summations run over the connected neighbours.
412: %%It is easy to check that the set of equations \eqref{ESWN/SDE/1} satisfies
413: %%the normalization condition \eqref{ESWN/SOCL/1}.
414: Clearly the set of equations \eqref{ESWN/SDE/1} is not closed
415: since it involves pair probabilities without describing their time evolution.
416: This follows from the stochastic process \eqref{ESWN/scheme}
417: where infection \eqref{ESWN/scheme/disease} proceeds \textit{via} $SI$ contact pairs.
418:
419: %%Dynamical equations for the pair probabilities can be easily derived,
420: %%%%The pair probabilities $\prob{AB}$ satisfy the normalization condition
421: %%%%\begin{equation}
422: %%%%\label{ESWN/SOCL/2}
423: %%%% \prob{SS}+\prob{II}+\prob{RR}%
424: %%%% +
425: %%%% 2%
426: %%%% \bigl[%
427: %%%% \prob{SI}+\prob{SR}+\prob{IR}%
428: %%%% \bigr]%
429: %%%% =1
430: %%%% ,
431: %%%%\end{equation}
432: %%%%and
433: %%their time evolution is given by:
434: %%\begin{subequations}
435: %%\label{ESWN/SDE/2}
436: %%\begin{align}
437: %%\label{ESWN/SDE/2/SS}
438: %% \difft{\prob{SS}}&=%
439: %% +2\,\eswnDemographicRate\:%
440: %% \bigl[%
441: %% \prob{SI}%
442: %% +%
443: %% \prob{SR}%
444: %% \bigr]
445: %% -2\,\eswnInfectionRate\sum_{\eswnNeighbourIndex}{\prob{{SS}{I_{\eswnNeighbourIndex}}}}%
446: %% ,
447: %% \\
448: %%\label{ESWN/SDE/2/II}
449: %% \difft{\prob{II}}&=%
450: %% +2\,\eswnInfectionRate\:
451: %% \Bigl[%
452: %% \prob{SI}%
453: %% +%
454: %% \sum_{\eswnNeighbourIndex}{\prob{{IS}{I_{\eswnNeighbourIndex}}}}%
455: %% \Bigr]
456: %% -2\,(\eswnDemographicRate+\eswnRecoveryRate)\:%
457: %% \prob{II}%
458: %% ,
459: %% \\
460: %%\label{ESWN/SDE/2/RR}
461: %% \difft{\prob{RR}}&=%
462: %% +2\,\eswnRecoveryRate\:%
463: %% \prob{IR}
464: %% -2\,\eswnDemographicRate\:%
465: %% \prob{RR}%
466: %% ,
467: %% \\
468: %%\label{ESWN/SDE/2/SI}
469: %% \difft{\prob{SI}}&=%
470: %% +\eswnDemographicRate\:%
471: %%%% \prob{I}
472: %% \bigl[%
473: %% \prob{SI}%
474: %% +%
475: %% \prob{II}%
476: %% +%
477: %% \prob{IR}%
478: %% \bigr]
479: %% -\eswnInfectionRate\:%
480: %% \Bigl[%
481: %% \sum_{\eswnNeighbourIndex}{\prob{{IS}{I_{\eswnNeighbourIndex}}}}%
482: %% -%
483: %% \sum_{\eswnNeighbourIndex}{\prob{{SS}{I_{\eswnNeighbourIndex}}}}%
484: %% \Bigr]
485: %% -(2\,\eswnDemographicRate+\eswnInfectionRate+\eswnRecoveryRate)\:%
486: %% \prob{SI}%
487: %% ,
488: %% \\
489: %%\label{ESWN/SDE/2/SR}
490: %% \difft{\prob{SR}}&=%
491: %% +\eswnDemographicRate\:%
492: %% \bigl[%
493: %% \prob{IR}%
494: %% +%
495: %% \prob{RR}%
496: %% -%
497: %% \prob{SR}%
498: %% \bigr]
499: %% +\eswnRecoveryRate\:%
500: %% \prob{SI}
501: %% -\eswnInfectionRate\:%
502: %% \sum_{\eswnNeighbourIndex}{\prob{{RS}{I_{\eswnNeighbourIndex}}}}%
503: %% ,
504: %% \\
505: %%\label{ESWN/SDE/2/IR}
506: %% \difft{\prob{IR}}&=%
507: %% +\eswnInfectionRate\:%
508: %% \sum_{\eswnNeighbourIndex}{\prob{{RS}{I_{\eswnNeighbourIndex}}}}
509: %% +\eswnRecoveryRate\:%
510: %% \prob{II}
511: %% -(2\,\eswnDemographicRate+\eswnRecoveryRate)\:%
512: %% \prob{IR}%
513: %% .
514: %%\end{align}
515: %%\end{subequations}
516: %%%%By substituting equations \eqref{ESWN/SDE/2} into the conservation laws
517: %%%%\begin{subequations}
518: %%%%\label{ESWN/SOCL/P2S}
519: %%%%\begin{align}
520: %%%% \prob{SS}+\prob{SI}+\prob{SR}&=\prob{S}
521: %%%% ,
522: %%%% \\
523: %%%% \prob{SI}+\prob{II}+\prob{IR}&=\prob{I}
524: %%%% ,
525: %%%% \\
526: %%%% \prob{SR}+\prob{IR}+\prob{RR}&=\prob{R}
527: %%%% ,
528: %%%%\end{align}
529: %%%%\end{subequations}
530: %%%%differentiated once and by noting that the pairs and triplets satisfy
531: %%%%\begin{equation}
532: %%%%\label{ESWN/SOCL/sT2sP}
533: %%%% \prob{SI}%
534: %%%% +%
535: %%%% \sum_{\eswnNeighbourIndex}{\prob{{SS}{I_{\eswnNeighbourIndex}}}}%
536: %%%% +%
537: %%%% \sum_{\eswnNeighbourIndex}{\prob{{IS}{I_{\eswnNeighbourIndex}}}}%
538: %%%% +%
539: %%%% \sum_{\eswnNeighbourIndex}{\prob{{RS}{I_{\eswnNeighbourIndex}}}}%
540: %%%% =%
541: %%%% \sum_{\eswnNeighbourIndexPrime}{\prob{{S}{I_{\eswnNeighbourIndexPrime}}}}%
542: %%%% ,
543: %%%%\end{equation}
544: %%%%it is easy to check that the set of equations \eqref{ESWN/SDE/2}
545: %%%%subject to the normalization condition \eqref{ESWN/SOCL/2} is consistent
546: %%%%with the equations for the singleton probabilities \eqref{ESWN/SDE/1} as required.
547: %%As expected,
548: %%the set of equations for the pair probabilities \eqref{ESWN/SDE/2}
549: %%is not closed as it includes triplet probabilities.
550:
551: %%Clearly, this inductive process may be repeated up to any arbitrary order $q$:
552: As a matter of fact,
553: the time evolution of the $q$-tuple probabilities is described
554: by a set of first order differential equations expressing
555: their time derivatives as linear combinations of $q$-tuple and $(q\!+\!1)$-tuple probabilities,
556: subject to a normalization condition.
557: In order to proceed,
558: the set of equations must be closed,
559: that is the $(q\!+\!1)$-tuple probabilities must be written in terms of $q$-tuple probabilities.
560: The `art' is to use closures that capture
561: key physical features of the system and are still manageable by symbolic or numerical-symbolic computation.
562: The results of a particular closure, or approximation,
563: may then be checked against rigorous results and/or stochastic simulations.
564:
565: For most closures the $(q\!+\!1)$-tuple probabilities are rational functions of the $q$-tuple probabilities,
566: appropriately normalized,
567: and thus the constrained set of first order differential equations may be replaced by an unconstrained set
568: where the time derivatives of independent $q$-tuple probabilities are expressed as
569: rational functions of these $q$-tuple probabilities.
570: Although the resulting sets of equations
571: are easily integrable by classical numerical methods
572: and admit polynomial systems as steady state equations,
573: their analysis remains cumbersome even at low order $q$.
574:
575: The simplest closure is the mean field approximation (\textsc{MFA}),
576: where the pairs ($2$-tuples) are assumed to be formed by
577: uncorrelated singletons ($1$-tuples):
578: \begin{equation}
579: \label{ESWN/UCS/approximation}
580: \sum_{\eswnNeighbourIndex}{\prob{{S}{I_{\eswnNeighbourIndex}}}}\approx%
581: \eswnDegree\:%
582: \prob{S}\prob{I}
583: .
584: \end{equation}
585: For the demographic \textsc{SIR} model the endemic equilibrium
586: (steady state)
587: is computed easily.
588: %%The endemic infective probability (order parameter) $\enprob{I}$ reads
589: %%\begin{equation}
590: %%\label{ESWN/UCS/eip}
591: %% \enprob{I}=%
592: %% \begin{cases}
593: %% 0&
594: %% \text{for $\eswnInfectionRate<\eswnCriticalInfectionRate$},\\
595: %% \frac{\eswnDemographicRate}{\eswnDegree}\:%
596: %% \frac{{\eswnInfectionRate}-{\eswnCriticalInfectionRate}}%
597: %% {\eswnInfectionRate\eswnCriticalInfectionRate}%
598: %% &
599: %% \text{for $\eswnCriticalInfectionRate\leqslant\eswnInfectionRate$}%
600: %% ,
601: %% \end{cases}
602: %%\end{equation}
603: %%where the endemic threshold
604: %%(or critical infection rate)
605: %%$\eswnCriticalInfectionRate$ is given by
606: %%\begin{equation}
607: %%\label{ESWN/UCS/cir}
608: %% \eswnCriticalInfectionRate=%
609: %%%% \frac{\eswnRecoveryRate+\eswnDemographicRate}{\eswnDegree}
610: %% \frac{1}{\eswnDegree}\:(\eswnRecoveryRate+\eswnDemographicRate)
611: %% .
612: %%\end{equation}
613: The mean-field endemic infective probabilities are plotted in Figure~\ref{fig/X/endemic}
614: as a function of the infection rate, at two values of $\eswnDemographicRate$.
615: For any value of $\eswnDemographicRate$,
616: the \textsc{MFA} predicts two different steady states:
617: at infection rates $\eswnInfectionRate$ smaller than
618: the critical infection rate $\eswnCriticalInfectionRate$
619: there is disease extinction,
620: %%(\mbox{$\enprob{I}=0$} or \mbox{$\enprob{S}=1$}),
621: while at infection rates $\eswnInfectionRate$ greater than
622: the critical infection rate $\eswnCriticalInfectionRate$
623: there is disease persistence,
624: \textit{i.e.} infected (and recovered) individuals coexist with susceptibles.
625: The two regimes are separated by the mean-field endemic threshold
626: that is plotted in Figure~\ref{fig/UPA/phasediagram} (dashed line).
627: %%We note that this threshold differs from the mean-field threshold of the \textsc{SIRS} model.
628:
629: \begin{figure}[b]
630: \begin{center}
631: \includegraphics[width=1.0\linewidth]{pamds-PhaseDiagram-UPA}%
632: \end{center}
633: \caption{%% see `pamds/meta/pamds-PhaseDiagram-UPA.mp.data.mbj'
634: Phase diagram for the \textsc{UPA}:
635: the no-coexistence phase and the coexistence phase
636: are separated by
637: the critical curve from
638: simulations (open circles),
639: the \textsc{MFA} (long dashed line),
640: and the \textsc{UPA} (thick solid line).
641: Within the coexistence phase,
642: at very low demographic rates
643: $\eswnDemographicRate$,
644: the \textsc{UPA} predicts
645: an oscillatory phase
646: as shown in the inset.
647: }%
648: \label{fig/UPA/phasediagram}
649: \end{figure}
650:
651: We anticipate that the results of the \textsc{MFA} will be accurate
652: when the demographic process of \eqref{ESWN/scheme} dominates over the infectious one
653: since in this regime pairs are continually broken
654: and thus the behaviour of each individual is essentially independent on that of the other ones.
655: The infection process governed by \textsl{Susceptible}-\textsl{Infective} contact pairs,
656: dominates in the opposite regime (\mbox{$\eswnDemographicRate\ll{1}$}),
657: relevant in the epidemiological context.
658: The appropriate mean field theory is then the uncorrelated pair approximation (\textsc{UPA}).
659:
660: The \textsc{UPA} is for pairs what the \textsc{MFA} is for singletons.
661: In the \textsc{UPA} triplets ($3$-tuples) are assumed to be formed
662: by uncorrelated pairs:
663: \begin{equation}
664: \label{ESWN/UCP/approximation}
665: \sum_{\eswnNeighbourIndex}{\prob{{AS}{I_{\eswnNeighbourIndex}}}}\approx%
666: (\eswnDegree-1)\:%
667: \frac{\prob{SA}\prob{SI}}{\prob{S}}
668: .
669: \end{equation}
670: The \textsc{UPA} is expected to outperform the \textsc{MFA} but,
671: in general,
672: its solution is not known in closed form.
673: For the demographic \textsc{SIR} model
674: the calculation of the phase diagram and the stability analysis
675: is still tractable by symbolic computation.
676: For lattices with coordination number \mbox{$\eswnDegree\!=\!4$}, the
677: phase diagram is plotted in Figure~\ref{fig/UPA/phasediagram}.
678: It is clear that the \textsc{UPA} is quantitatively superior to the \textsc{MFA}
679: when compared with the results of simulations (open circles).
680: Both the \textsc{MF} and the \textsc{UP} approximations
681: of the \mbox{$\eswnDegree\!=\!4$} demographic \textsc{SIR} model predict
682: a finite critical infection rate at $\eswnDemographicRate=0$,
683: while the simulations indicate
684: that $\eswnCriticalInfectionRate$ will diverge as $\eswnDemographicRate$ tends to $0$.
685: %%This is compatible with the results of computer
686: %%simulations of Figure~\ref{fig/UPA/phasediagram}
687: %%performed along rings of size \mbox{$\eswnN\!=\!2^{18}$}.
688:
689:
690: However,
691: the \textsc{SIRS} and the demographic \textsc{SIR} models
692: are different at low
693: (but finite)
694: demographic rates $\eswnDemographicRate$.
695: In the demographic \textsc{SIR} model the mechanism
696: for the renewal of susceptibles is totally random by contrast to
697: the mechanism of the \textsc{SIRS} model.
698: In our model susceptibles are born anywhere on the lattice while
699: in the \textsc{SIRS} model only previously infected sites loose immunity.
700: %%This suggests that
701: %%the demographic rate $\eswnDemographicRate$ is much more effective in mixing
702: %%(and thus in destroying coupling),
703: %%allowing the existence of a finite critical infection rate,
704: %%even in one-dimensional demographic \textsc{SIR} models.
705:
706: We note that
707: the randomizing effect of the demographic \textsc{SIR} mechanism
708: for the renewal of susceptibles is reminiscent
709: of the randomizing effect of shortcuts in small-world networks of the
710: Watts and Strogatz type \cite{WattsStrogatz1998,Hastings2003}
711: where correlations are destroyed
712: and an effective mixing of the population is achieved,
713: with (drastic) consequences on the phase diagram.
714:
715: Finally, it is worth noticing that the \textsc{UPA} also predicts the existence
716: of an oscillatory phase within the survival or coexistence phase
717: (\textit{i.e.}, to the right of $\eswnCriticalInfectionRate(0)$),
718: for small values of $\eswnDemographicRate$
719: (Figure~\ref{fig/UPA/phasediagram}).
720: The same is true for the \textsc{UPA} of process \eqref{ESWN/SIRS/scheme}
721: on the square lattice.
722: %%This behaviour has to be confirmed and investigated further.
723: This behaviour will be difficult to identify in stochastic simulations,
724: since it may be blurred by large fluctuations and stochastic extinctions.
725:
726: \begin{figure}[t]
727: \begin{center}
728: \includegraphics[width=1.0\linewidth]{pamds-PhaseDiagram-CPA}%
729: \end{center}
730: \caption{%% see `pamds/meta/pamds-PhaseDiagram-CPA.mp.data.mbj'
731: Isoparametric phase diagrams for the correlated model:
732: the no-coexistence and coexistence phases
733: are separated by the critical curve from
734: simulations (open circles),
735: the \textsc{MFA} (long dashed line),
736: the \textsc{UPA}
737: % or for a vanishing closed form parameter
738: (dashed dotted line)
739: and the correlated model for different
740: % closed form parameters
741: $\eswnClosedFormParameter$
742: (solid lines).
743: For
744: % the remarkable closed form parameter
745: $\eswnRemarkableClosedFormParameter\!\approx\!{0.3807}$
746: % \footnotemark[1]
747: (bold solid line)
748: the critical infection rate $\eswnCriticalInfectionRate$ tends asymptotically to infinity
749: when the demographic rate $\eswnDemographicRate$ vanishes.
750: Closed form parameters $\eswnClosedFormParameter$
751: from left to right:
752: $\tfrac{1}{4}$,
753: $\eswnClosedFormParameter^{*}$,
754: $\tfrac{1}{2}$,
755: $\tfrac{5}{8}$,
756: $\tfrac{3}{4}$.
757: }%
758: \label{fig/CLM/phasediagram/isoparametric}
759: \end{figure}
760:
761: \section{Correlated pair approximations}\label{sec/CPA}
762: In order to construct more realistic pair approximations,
763: we have investigated closure procedures inspired
764: by the geometrical structure of the lattice.
765:
766: Within this perspective and as far as social triplets are concerned,
767: the ring of degree \mbox{$\eswnDegree\!=\!4$}
768: and the triangular lattice \mbox{($\eswnDegree\!=\!6$)}
769: are propitious networks
770: since their nearest-neighbour triplets split into two distinct classes:
771: `chain-like' (open) and `loop-like' (closed) triplets.
772: %%On the ring of degree \mbox{$\eswnDegree\!=\!4$}
773: %%the probability of finding open and closed triplets are equal,
774: %%whereas on the regular triangular lattice they are different,
775: %%$3/5$ and $2/5$ respectively.
776: A very naive idea is to take into account the two classes of triplets
777: and to use the probability $\eswnClosedFormParameter$ and $1-\eswnClosedFormParameter$
778: of finding respectively a `loop-like' triplet and a `chain-like' triplet
779: as a parameter to be fitted to simulation results.
780: Thus triplets are assumed to be formed
781: either of uncorrelated (chained) pairs
782: or of correlated (looped) pairs
783: \cite{VanBaalen2000}:
784: \begin{equation}
785: \label{ESWN/CPL/approximation}
786: \begin{split}
787: &\qquad%
788: \sum_{\eswnNeighbourIndex}{\prob{{AS}{I_{\eswnNeighbourIndex}}}}\approx%
789: \\
790: &%
791: \begin{cases}
792: (\eswnDegree-1)\,%
793: \Bigr[%
794: \bigr(1-\eswnClosedFormParameter\bigl)\,%
795: \tfrac{\prob{SA}\prob{SI}}{\prob{S}}%
796: +%
797: \,%
798: \eswnClosedFormParameter\:%
799: \tfrac{\prob{AI}\prob{SA}\prob{SI}}{\prob{A}\prob{S}\prob{I}}%
800: \Bigl]%
801: &\\
802: %%%% \tfrac{\prob{SA}\prob{SI}}{\prob{S}}%
803: %%%% \Bigr[%
804: %%%% \bigr(1-\eswnClosedFormParameter\bigl)%
805: %%%% +%
806: %%%% \,%
807: %%%% \eswnClosedFormParameter\,%
808: %%%% \tfrac{\prob{AI}}{\prob{A}\prob{I}}%
809: %%%% \Bigl]%
810: % &%
811: %%%% \text{if ${A}\neq{I}$}%
812: \qquad\text{if ${A}\in\{S,R\}$}%
813: ,
814: &\\
815: (\eswnDegree-1)\,%
816: \prob{SI}%
817: -%
818: \sum_{\eswnNeighbourIndex}{%
819: \bigr[%
820: \prob{{SS}{I_{\eswnNeighbourIndex}}}%
821: \!+\!%
822: \prob{{RS}{I_{\eswnNeighbourIndex}}}%
823: \bigl]%
824: }
825: &\\
826: %%%% \Bigl[%
827: %%%% \sum_{\eswnNeighbourIndex}{\prob{{SS}{I_{\eswnNeighbourIndex}}}}%
828: %%%% +%
829: %%%% \sum_{\eswnNeighbourIndex}{\prob{{RS}{I_{\eswnNeighbourIndex}}}}%
830: %%%% \Bigr]%
831: \qquad\text{if ${A}={I}$}%
832: .
833: &
834: \end{cases}
835: \end{split}
836: \end{equation}
837: %%\begin{equation}
838: %%\label{ESWN/CPL/approximation}
839: %% \sum_{\eswnNeighbourIndex}{\prob{{AS}{I_{\eswnNeighbourIndex}}}}\!\approx\!%
840: %% \begin{cases}
841: %% (\eswnDegree-1)\,%
842: %% \Bigr[%
843: %% \bigr(1-\eswnClosedFormParameter\bigl)\,%
844: %% \tfrac{\prob{SA}\prob{SI}}{\prob{S}}%
845: %% +%
846: %% \,%
847: %% \eswnClosedFormParameter\:%
848: %% \tfrac{\prob{AI}\prob{SA}\prob{SI}}{\prob{A}\prob{S}\prob{I}}%
849: %% \Bigl]%
850: %%%% \tfrac{\prob{SA}\prob{SI}}{\prob{S}}%
851: %%%%%% \Bigr[%
852: %%%%%% \bigr(1-\eswnClosedFormParameter\bigl)%
853: %%%%%% +%
854: %%%%%% \,%
855: %%%%%% \eswnClosedFormParameter\,%
856: %%%%%% \tfrac{\prob{AI}}{\prob{A}\prob{I}}%
857: %%%%%% \Bigl]%
858: %% &%
859: %%%%%% \text{if ${A}\neq{I}$}%
860: %% \text{if ${A}\in\{S,R\}$}%
861: %% ,
862: %% \\
863: %% (\eswnDegree-1)\,%
864: %% \prob{SI}%
865: %% -%
866: %% \sum_{\eswnNeighbourIndex}{%
867: %% \bigr[%
868: %% \prob{{SS}{I_{\eswnNeighbourIndex}}}%
869: %% \!+\!%
870: %% \prob{{RS}{I_{\eswnNeighbourIndex}}}%
871: %% \bigl]%
872: %% }
873: %%%%%% \Bigl[%
874: %%%%%% \sum_{\eswnNeighbourIndex}{\prob{{SS}{I_{\eswnNeighbourIndex}}}}%
875: %%%%%% +%
876: %%%%%% \sum_{\eswnNeighbourIndex}{\prob{{RS}{I_{\eswnNeighbourIndex}}}}%
877: %%%%%% \Bigr]%
878: %% &%
879: %% \text{if ${A}={I}$}%
880: %% .
881: %% \end{cases}
882: %%\end{equation}
883:
884: The demographic \textsc{SIR} version of the \textsc{CPA}
885: \eqref{ESWN/CPL/approximation}
886: is amenable by cumbersome numerical-symbolic computation although some
887: interesting results may be obtained by symbolic computation.
888: The phase diagrams
889: are shown in Figure~\ref{fig/CLM/phasediagram/isoparametric}.
890: We find that, as $\eswnClosedFormParameter$ increases
891: from $0$ to $\eswnRemarkableClosedFormParameter\!\approx\!{0.3807}$%
892: \footnote{%
893: the real solution of the cubic equation
894: $27\eswnClosedFormParameter^{3}-18\eswnClosedFormParameter^{2}+87\eswnClosedFormParameter-32=0$.
895: }%
896: ,
897: keeping \mbox{$\eswnDegree\!=\!4$} fixed,
898: the \textsc{CPA} phase diagrams interpolate
899: between the \textsc{UPA} behaviour
900: and typical one-dimensional phase diagrams with $\eswnCriticalInfectionRate(0)=\infty$.
901: At $\eswnRemarkableClosedFormParameter$
902: the critical infection rate $\eswnCriticalInfectionRate$ tends asymptotically
903: to infinity as the demographic rate $\eswnDemographicRate$ vanishes.
904: Inspection of Figure~\ref{fig/CLM/phasediagram/isoparametric} also shows that
905: the closed form parameter $\eswnClosedFormParameter$ cannot be constant
906: if a quantitative description of the global phase diagram is required.
907: If we allow $\eswnClosedFormParameter$ to depend on $\eswnDemographicRate$,
908: reasonable descriptions of the endemic equilibria
909: (Figure~\ref{fig/X/endemic})
910: and of the global phase diagram
911: (Figure~\ref{fig/CLM/phasediagram/isoparametric})
912: are obtained.
913: For the \textsc{SIRS} model \eqref{ESWN/SIRS/scheme} on the square lattice,
914: a \textsc{CPA} obtained by fitting $\eswnClosedFormParameter$ to $\eswnCriticalInfectionRate(0)$
915: will improve the results of the \textsc{UPA} used
916: in \cite{JooLebowitz2004}
917: to describe the behaviour of the system at low values of $\eswnDemographicRate$.
918:
919: \section{Discussion}\label{sec/discussion}
920: We have proposed a simple \textsc{CPA} that
921: was shown to provide a reasonable approximation
922: to the behaviour of stochastic models that are relevant in epidemiology
923: ---
924: the agreement against simulation data
925: being far better than \textsc{MFA} and \textsc{UPA}
926: with a suitable choice of the parameters.
927: The resulting equations of evolution may be used to approximate phase diagrams,
928: as well as steady state and dynamical behaviours of the associated stochastic models.
929: The \textsc{CPA} takes into account some of the effects of the local lattice structure
930: and yields a clear alternative to heavy stochastic simulations.
931:
932: One of the directions of future work includes the development of \textsc{CPA}s,
933: along the lines of the present work,
934: to account for the local (lattice like) structure of a class of
935: complex networks,
936: such as the Watts and Strogatz small-world networks,
937: that have been shown to be relevant in epidemiological contexts
938: \cite{Verdasca2005}.
939:
940: \begin{acknowledgement}
941: Financial supports from
942: the Foundation of the University of Lisbon,
943: under contract
944: BPD-CI/01/04,
945: and from
946: the Portuguese Foundation for Science and
947: Technology (FCT),
948: under contracts
949: POCTI/ESP/44511/2002
950: and
951: POCTI/ISFL/2/618,
952: are gratefully acknowledged.
953: \end{acknowledgement}
954:
955: \bibliographystyle{apsrev}
956: %%% \bibliography{eswn,phyabbrev}
957: \bibliography{pamds}
958: %% Bibliography:
959:
960: \end{document}
961: %%
962: %% End of the file `pamds.tex'.
963: