1: %\documentclass[twocolumn,showpacs,aps,floatfix,superscriptaddress]{revtex4}
2: %\documentclass[twocolumn,aps,showpacs,amsmath,amssymb,floatfix,superscriptaddress,letterpaper]{revtex4}
3: %\documentclass[twocolumn,aps,showpacs,amsmath,amssymb,floatfix,superscriptaddress,prl]{revtex4}
4: %\documentclass[aps,amsmath,amssymb,showpacs,floatfix]{revtex4}
5: %\documentclass[twocolumn,aps,showpacs,floatfix]{amsart}
6: %\usepackage{graphicx}% Include figure files
7: %\usepackage{hyperref}
8: %\usepackage{latexsym}
9: %\usepackage{ifsym}
10: \documentclass[twocolumn,aps,showpacs,prl,floatfix]{revtex4}
11: \usepackage{amsmath,amssymb,eucal,graphicx,hyperref}
12: \usepackage{dcolumn}% Align table columns on decimal point
13: \usepackage{bm}% bold math
14: \usepackage{array}
15: \usepackage{float}
16: \usepackage{supertabular}
17: \usepackage{longtable}
18: \usepackage{txfonts}
19: \begin{document}
20:
21: \title{Evolutionary dynamics on degree-heterogeneous graphs}
22: \author{T. Antal} \author{S. Redner} \author{V. Sood} \affiliation{Center of Polymer
23: Studies and Department of Physics, Boston University, Boston,
24: Massachusetts, 02215 USA}
25:
26: \begin{abstract}
27:
28: The evolution of two species with different fitness is investigated on
29: degree-heterogeneous graphs. The population evolves either by one
30: individual dying and being replaced by the offspring of a random neighbor
31: (voter model (VM) dynamics) or by an individual giving birth to an
32: offspring that takes over a random neighbor node (invasion process (IP)
33: dynamics). The fixation probability for one species to take over a
34: population of $N$ individuals depends crucially on the dynamics and on the
35: local environment. Starting with a single fitter mutant at a node of
36: degree $k$, the fixation probability is proportional to $k$ for VM dynamics
37: and to $1/k$ for IP dynamics.
38:
39: \end{abstract}
40:
41: \pacs{87.23.Kg, 05.40.-a, 02.50.-r, 89.75.-k}
42: \maketitle
43:
44: In this letter, we investigate the likelihood for fitter mutants to
45: overspread an otherwise uniform population on heterogeneous graphs by
46: evolutionary dynamics. Such a process underlies epidemic propagation
47: \cite{epidemics}, emergence of fads \cite{fads}, social cooperation
48: \cite{coop}, or invasion of an ecological niche by a new species
49: \cite{moran::MP,invasion,lieberman::EDG}. At each update event, two
50: individuals from a total population $N$ are chosen at random. One individual
51: replicates while the other dies and is replaced by the newly-born offspring,
52: so that $N$ remains constant. A selective advantage, or fitness, also exists
53: in which each each individual may be a unit-fitness genotype ${\bf 1}$ or
54: genotype ${\bf 0}$ with lower fitness $1-s$, with $0<s<1$. These fitnesses
55: determine the replication or death rates of each individual. This selective
56: advantage leads to a dynamical competition in which selection dominates for
57: large populations, while random genetic drift \cite{kimura::NTE,ewens::MPG}
58: occurs for small populations or weak selection.
59:
60: We consider three evolutionary models, distinguished by the order in which a
61: pair individuals replicate and die:
62:
63: \noindent{\tt Biased Link Dynamics (LD):} A link is selected at random. If
64: the individuals at the link ends are different, one of them is designated as
65: the ``donor'' with probability proportional to its fitness. The replicate of
66: the donor then replaces the other individual: ${\bf 10} \to {\bf 11}$ with
67: probability $1/2$ while ${\bf 10}\to {\bf 00}$ with probability (1-s)/2
68: (Fig.~\ref{update}).
69:
70: \noindent{\tt Biased Voter Model (VM):} An individual dies with probability
71: inversely proportional to its fitness, and is then replaced by the offspring
72: of a randomly-chosen neighbor. Equivalently, death occurs randomly and
73: replacement is proportional to the fitness of the donor. We implement the VM
74: by updating a randomly-chosen genotype {\bf 0} with probability 1, while the
75: fitter genotype {\bf 1} is updated with a probability $1-s$. Each individual
76: in this death-first/birth-second process can equivalently be viewed as a
77: voter that adopts the opinion of a randomly-selected neighbor
78: \cite{liggett,krapivsky::VM,sood::VM}.
79:
80: \noindent{\tt Biased Invasion Process (IP):}
81: In this birth-first/death-second process, a randomly-chosen individual
82: replicates with probability proportional to its fitness, and its offspring
83: then replaces an individual at a randomly-chosen neighboring node
84: \cite{ewens::MPG,moran::MP,invasion}.
85:
86: \begin{figure}[ht]
87: \vspace*{0.cm}
88: \includegraphics*[width=0.35\textwidth]{update}
89: \caption{Update illustration for two specific nodes. Genotypes {\bf 0} and
90: {\bf 1} are denoted by $\medcirc$ and $\medbullet$ respectively. Shown are the
91: possible transitions and their respective relative rates due to the
92: interaction of two nodes across a link for LD, VM, and IP dynamics.}
93: \label{update}
94: \end{figure}
95:
96: One genotype ultimately replacing all other genotypes in the population is
97: termed fixation. An important, and easily checked fact is that these
98: evolutionary models are equivalent on degree-regular graphs; moreover, as we
99: will show, the fixation probability in LD can be obtained exactly,
100: independent of the underlying graph. However, essential differences arise on
101: degree-heterogeneous networks \cite{sood::VM,SEM,C05} that may lead to an
102: enhancement of the fixation probability, as discovered previously for the IP
103: \cite{lieberman::EDG}. Here we cast LD, VM, and IP on degree-heterogeneous
104: graphs within the same unifying framework to understand the interplay between
105: selection and random drift on the fixation probability. By this approach, we
106: show that on degree-heterogeneous graphs the best strategy to reach fixation
107: with VM dynamics is for the fitter genotype to be on high-degree nodes.
108: Conversely, for IP dynamics, it is best for the fitter genotype to be on
109: low-degree nodes.
110:
111: % LD, IP, and VM dynamics are equivalent on degree-regular graphs \cite{SEM};
112: % moreover the fixation probability in LD can be obtained exactly, independent
113: % of the underlying graph. On degree-heterogeneous networks, however, the
114: % three dynamics lead to different outcomes. For the IP, in particular,
115: % special graphs were identified that gave an enhanced fixation probability
116: % compared to that on regular graphs \cite{lieberman::EDG}. In a similar vein,
117: % we will show that fixation for biased VM dynamics has a rich behavior that
118: % arises from the competition between selection and random drift.
119:
120: We first study the evolution in VM dynamics. We symbolically represent the
121: state of the system by $\eta$. In an elemental time interval $\delta t$ we
122: choose a random node $x$. If the genotype at this node at time $t$, denoted
123: as $\eta^t(x)$, equals ${\bf 0}$, then node $x$ is updated by choosing a
124: random neighbor $y$ and setting $\eta^{(t+\delta t)}(x)=\eta^t(y)$
125: (Fig.~\ref{update}). However if $\eta^t(x) = {\bf 1}$, the VM update is
126: implemented with probability $1-s$. This update rule can be written as
127: \begin{eqnarray}
128: \label{trans::voter}
129: {\bf P}[\eta\!\to\!\eta_x]=
130: \sum_y \frac{A_{x y}}{Nk_x}\Big\{[1\!-\!\eta(x)]\eta(y)
131: +(1\!-\!s)\eta(x)[1\!-\!\eta(y)]\Big\}\!,
132: \end{eqnarray}
133: % \begin{eqnarray}
134: % \label{trans::voter}
135: % {\bf P}[\eta\to\eta_x]&=&\frac{1}{N}
136: % \sum_y \frac{A_{x y}}{k_x}\Big\{[1-\eta(x)]\eta(y)\nonumber \\
137: % &~&~~~~~~~~~+(1-s)\eta(x)[1-\eta(y)]\Big\}\,,
138: % \end{eqnarray}
139: where $\eta_x$ denotes the state obtained from $\eta$ by changing only the
140: genotype at node $x$. The first term describes the update step for the case
141: where $(\eta(x),\eta(y))= ({\bf 0},{\bf 1})$ and $x, y$ are connected.
142: Each of the nearest neighbors $y$ of $x$ may be selected with probability
143: $A_{xy}/k_x$. Here $A_{xy}$ is the adjacency matrix whose elements equal 1
144: if $xy$ are connected and zero otherwise. The second term in
145: Eq.~\eqref{trans::voter} is explained analogously.
146:
147: For degree-heterogeneous graphs, the density $\rho_k$ of genotype {\bf 1} at
148: nodes of degree $k$ increases by $1/N_k$ with probability ${\bf F}_k(\eta)$ and
149: decreases by $1/N_k$ with probability ${\bf B}_k(\eta)$ in an elemental update,
150: where
151: \begin{equation}
152: \label{FBkdef}
153: \begin{split}
154: {\bf F}_k(\eta) &= \frac{1}{k N}\mathop{{\sum}'}_{x y~} A_{x y} [1-\eta(x)]\eta(y)\\
155: {\bf B}_k(\eta) &= \frac{1-s}{k N}\mathop{{\sum}'}_{x y~}A_{x y} \eta(x)[1-\eta(y)]
156: \end{split}
157: \end{equation}
158: are the forward ({\bf 0} $\to$ {\bf 1}) and backward ({\bf 1} $\to$ {\bf 0})
159: evolution rates. The primes on the sums denote the restriction that the
160: degree of nodes $x$ equals $k$. The sum over all $k$ then gives the total
161: transition rate of Eq.~\eqref{trans::voter}. We seek the fixation
162: probability $\Phi$ to the state consisting entirely of genotype {\bf 1} as a
163: function of the initial densities of {\bf 1}. This probability obeys the
164: backward Kolmogorov equation $G\Phi=0$ \cite{karlin}, subject to the boundary
165: conditions $\Phi(0)=0$ and $\Phi(1)=1$. In the diffusion approximation, the
166: generator $G$ of this equation may be expressed as a sum of the changes in
167: $\rho_k$ over all $k$,
168: \begin{equation}
169: \label{gendef}
170: G = \frac{1}{\delta t}\sum_k
171: \left[ \delta\rho_k ({\bf F}_k\!-\!{\bf B}_k)\partial_k \!+\! \frac{(\delta\rho_k)^2}{2}({\bf F}_k
172: \!+\!{\bf B}_k)\partial_k^2\right]\,,
173: \end{equation}
174: with $\delta\rho_k = 1/N_k = 1/(N\, n_k)$ the change in $\rho_k$ in a single
175: update of a node of degree $k$, and
176: $\partial_k\equiv\frac{\partial}{\partial\rho_k}$.
177:
178: For the special case of degree-regular graphs, where $k_x=k$ for all nodes,
179: both sums in Eq.~\eqref{FBkdef} count the total number $\alpha$ of active
180: links between different genotypes
181: \begin{equation}
182: \alpha = \frac{1}{N \mu_1} \sum_{x,y} A_{xy} \eta(x) [1-\eta(y)] \,,
183: \end{equation}
184: where the moments of the degree distribution are defined by $\mu_n \equiv
185: N^{-1} \sum_x k_x^n = \sum_k k^n n_k$. The generator thus reduces to
186: \begin{equation}
187: \label{diffusion::regular}
188: G = \alpha\left[s\partial_{\rho} + \frac{1}{N}(1-\frac{s}{2})\partial_{\rho}^2\right]\,,
189: \end{equation}
190: where we use $\delta\rho = \delta t = 1/N$. In this form, the convection and
191: diffusion terms differ by a factor $\mathcal{O}(s N)$. Thus selection
192: dominates when the population $N$ is larger than $\mathcal{O}(1/s)$, while
193: random genetic drift is important otherwise.
194:
195: Notice that the probability of increasing the density of genotype {\bf 1} at
196: each update is a factor $1/(1-s)$ larger than the probability of decreasing
197: the density. By its construction, this same bias arises for LD on general
198: networks. As a consequence of this bias, the evolutionary process underlying
199: fixation is the same as the absorption of a uniformly biased random walk in a
200: finite interval \cite{karlin}, from which the fixation probability is
201: \begin{equation}
202: \label{fixation::regular}
203: \Phi(\rho) = \frac{1-(1-s)^{N\rho}}{1-(1-s)^{N}}\to \frac{1-e^{-sN\rho
204: /(1-s/2)}}{1-e^{-sN/(1-s/2)}}\,.
205: \end{equation}
206: The former is the exact discrete solution of $G\Phi=0$ on a finite network,
207: while the latter continuum limit represents the solution to $G\Phi=0$ in the
208: diffusion approximation. These results apply for all three models on
209: degree-regular graphs and for LD on general graphs.
210:
211: For degree-heterogeneous graphs, the conserved quantity for neutral dynamics
212: ($s=0$) is the average degree-weighted density $\omega_1$
213: \cite{sood::VM,SEM}, where the degree-weighted moments are
214: \begin{equation}
215: \omega_{n} = \frac{1}{N\mu_{n}} \sum_x k_x^n \eta(x) = \frac{1}{\mu_{n}} \sum_k k^n n_k \rho_k\,,
216: \end{equation}
217: while the overall density $\rho$ of genotype {\bf 1} is no longer conserved.
218: The existence of this new conservation law suggests that we study the time
219: evolution of the expectation value of $\omega_1$ which we henceforth denote
220: as $\omega$ for notational simplicity. Since $\omega(\eta_x) = \omega(\eta)
221: + k_x(1-2\eta(x))/\mu_1 N$,
222: \begin{eqnarray}
223: \label{evolution::omega::general}
224: \partial_t\omega &=& \frac{1}{\delta t}\sum_x
225: \left[\omega(\eta_x) - \omega(\eta)\right]{\bf P}[\eta\to\eta_x]\nonumber\\
226: &=& \frac{s}{\mu_1 N}\sum_{x,y}A_{x y} \eta(x)(1-\eta(y)) = s \alpha\,.
227: \end{eqnarray}
228: Notice that $\omega$ is conserved in the absence of selection ($s=0$) a
229: feature that ultimately stems from the update rate being inversely
230: proportional to node degree [Eq.~\eqref{trans::voter}]. To evaluate the
231: expression in Eq.~\eqref{evolution::omega::general} we make the mean-field
232: assumption that the degrees of connected nodes in the graph are uncorrelated.
233: Thus we replace the elements of the adjacency matrix by their expected
234: values, $A_{x y} = k_x k_y/\mu_1 N$. This assumption simplifies
235: Eq.~(\ref{evolution::omega::general}) to $\partial_t\omega =
236: s\omega(1-\omega)$, with solution $\omega(t)^{-1} =
237: 1-[1-\omega(0)^{-1}]e^{-st}$.
238:
239: For uncorrelated graphs, Eqs.~\eqref{FBkdef} simplify to
240: \begin{eqnarray}
241: \label{update::MR}
242: {\bf F}_k(\eta) = n_k \omega (1-\rho_k), \quad {\bf B}_k(\eta) = (1-s) n_k (1-\omega) \rho_k\,.
243: \end{eqnarray}
244: Thus the time evolution of the expectation value of $\rho_k$ is
245: \begin{equation}
246: \label{evolution::rhok::MR}
247: \partial_t\rho_k = \frac{\delta\rho_k({\bf F}_k - {\bf B}_k)}{\delta t}
248: = \omega-\rho_k + s(1-\omega)\rho_k\,.
249: \end{equation}
250: To solve this equation we combine it with $\partial_t\omega =
251: s\omega(1-\omega)$ to give $\partial_t(\omega-\rho_k) =
252: -(\omega-\rho_k)(1-s(1-\omega))$, with solution
253: \begin{equation}
254: \label{evolution::rhok::MR::solution}
255: \omega(t)\!-\!\rho_k(t) = e^{-t}[\omega(0)\!-\!\rho_k(0)]\{\omega(0) + [1\!-\!\omega(0)]e^{-s t}\}\,.
256: \end{equation}
257: For small selective advantage ($s\ll 1$), this equation involves two
258: distinct time scales. On a time scale of order one, all the $\rho_k$
259: become equal to $\omega$, whereas the evolution of $\omega$ occurs on a
260: longer time scale of order $s^{-1}\gg 1$ (Fig.~\ref{traj}).
261:
262: \begin{figure}[ht]
263: \vspace*{0.cm}
264: \includegraphics*[width=0.405\textwidth]{traj}
265: \caption{Moments of the {\bf 1} density in the biased VM and biased IP on a
266: network of $10^4$ nodes with a power-law degree distribution $n_k \sim
267: k^{-\nu}$ ($\nu=2.5$), and no correlations between node degrees. Nodes
268: with degree larger than the mean degree are initialized to {\bf 1} while
269: all other nodes are {\bf 0}. For the VM, $s=8.5 \times 10^{-4}$, while for
270: the IP, $s=10^{-4}$. }
271: \label{traj}
272: \end{figure}
273:
274: We now determine the fixation probability simply by replacing the $\rho_k$ by
275: $\omega$ in the forward and backward rates {\bf F} and {\bf B} in
276: Eqs.~\eqref{update::MR}. In a similar vein, we replace the derivative
277: $\partial_k$ by $(k n_k/\mu_1)\partial_{\omega}$ \cite{sood::VM}. Then the
278: generator in Eq.~\eqref{gendef} becomes
279: \begin{eqnarray}
280: \label{Kolmogorov::MR}
281: G &=& s\sum_k\left(\frac{k n_k}{\mu_1}\right)\omega(1-\omega)\partial_{\omega}\nonumber\\
282: &+& \frac{1}{N}\left(1-\frac{s}{2}\right)\sum_k\left(\frac{k^2 n_k}{\mu_1^2}\right)
283: \omega(1-\omega)\partial^2_{\omega}\nonumber\\
284: &=& \omega(1-\omega)\left[ s\partial_{\omega} +
285: \frac{\mu_2}{\mu_1^2 N}\left(1-\frac{s}{2}\right)\partial^2_{\omega}\right]\,.
286: \end{eqnarray}
287: Apart from an overall constant for the time scale, this generator is
288: identical to that of degree-regular graphs \eqref{diffusion::regular} if we
289: replace $N$ by an effective population size $N_{\rm eff} \equiv
290: N\mu_1^2/\mu_2$. For a network of $N$ nodes with a power-law degree
291: distribution, $n_k\propto k^{-\nu}$, $N_{\rm eff}$ scales as \cite{sood::VM}
292: \begin{equation}
293: \label{Neff}
294: N_{\rm eff} \equiv \frac{\mu_1^2 N}{\mu_2}\sim
295: \begin{cases}
296: N & \nu>3\,;\cr
297: N^{(2\nu-4)/(\nu-1)} & 2<\nu<3\,;\cr
298: \mathcal{O}(1) & \nu<2\,,
299: \end{cases}
300: \end{equation}
301: with logarithmic corrections for $\nu=2$ and $\nu=3$. Thus $N_{\rm eff}$
302: becomes much less than $N$ when $\mu_2$ diverges; this occurs when $\nu<3$.
303: A similar change in the effective size of the population is observed for
304: biological species evolving in a spatially heterogeneous environment
305: \cite{moran::MP,barton::1997}.
306:
307: The solution to $G\Phi=0$, with $G$ given by Eq.~\eqref{Kolmogorov::MR} is
308: \begin{equation}
309: \label{fixation::voter}
310: \Phi(\omega) = \frac{1-e^{-sN_{\rm eff}\omega /(1-s/2)\,}}{1-e^{-sN_{\rm eff}/(1-s/2)}}\,.
311: \end{equation}
312: Our numerical data for the fixation probability shows both excellent scaling
313: and agreement with this functional form for $\Phi$ (Fig.~\ref{surv-scaled}).
314: Eq.~\eqref{fixation::voter} also provides the fixation
315: probability when the system starts with a single mutant at a node of degree
316: $k$:
317: \begin{equation}
318: \label{fixation::voter-k}
319: \Phi_1 =
320: \begin{cases}
321: k/(N\mu_{1}) & s\ll 1/N_{\rm eff}\,; \\
322: sk\mu_1/\mu_2 & 1/N_{\rm eff} \ll s\ll 1\,.
323: \end{cases}
324: \end{equation}
325: The crucial feature is that the fixation probability of a single fitter
326: mutant is proportional to the degree of the node that it initially occupies
327: (Fig.~\ref{surv-one}). Notice also that because the relative effect of
328: selection versus random genetic drift is determined by the variable
329: combination $sN_{\rm eff}$, random genetic drift can be important for much
330: larger populations compared to the case of degree-regular graphs. In fact,
331: for a power-law graph with $\nu<2$, random genetic drift prevails for all
332: population sizes.
333:
334: \begin{figure}[ht]
335: \vspace*{0.cm}
336: \includegraphics*[width=0.405\textwidth]{sdep}
337: \caption{Scaling plot of fixation probabilities for VM (filled) and IP
338: dynamics (open symbols).
339: %Here $N_{\rm eff} =\mu_1^2 N/\mu_2$ for the VM
340: % and $N_{\rm eff}= N$ for IP.
341: Data are for degree-uncorrelated graphs with $N=10^3$, $\mu_1 = 8$, and
342: degree distribution exponent $\nu = 2.1$ ($\medcirc$), $ 2.5$
343: ($\triangle$), or $3.0$ ($\square$). Initially each node is a mutant with
344: probability 1/2, ($\omega_1=\omega_{-1}=1/2$). The curve corresponds to
345: Eqs.~\eqref{fixation::voter} or \eqref{fixation::invasion}.}
346: \label{surv-scaled}
347: \end{figure}
348:
349: \begin{figure}[ht]
350: \vspace*{0.cm}
351: \centerline{\includegraphics*[width=0.45\textwidth]{kdep}}
352: \caption{Fixation probability of a single mutant initially at a node of
353: degree $k$ on an uncorrelated power-law degree distributed ($n_k \sim
354: k^{-\nu}, ~\nu=2.5$) graph with $N=10^3$ and $\mu_1 = 8$. The empty
355: symbols correspond to IP dynamics with $s = 0.004$ ($\square$), $s = 0.008$
356: ($\medcirc$) and $s = 0.016$ ($\triangle$); the filled symbols correspond
357: to VM dynamics with $s = 0.01$, ($\blacksquare$), $s = 0.02$ ($\medbullet$)
358: and $s=0.08$ ($\blacktriangle$). The solid lines, with slopes $+1$ and
359: $-1$, correspond to the second of Eqs.~\eqref{fixation::voter-k} and
360: \eqref{fixation::invasion-k}.}
361: \label{surv-one}
362: \end{figure}
363:
364: We now study fixation in the complementary biased invasion process. Here a
365: randomly-selected individual reproduces with probability proportional to its
366: fitness; hence the transition probability is
367: \begin{eqnarray}
368: \label{trans::invasion}
369: {\bf P}[\eta\!\to\!\eta_x]\!=\!
370: \sum_y\! \frac{A_{x y}}{Nk_y}\Big\{[1\!-\!\eta(x)]\eta(y)
371: \!+\!(1\!-\!s)\eta(x)[1\!-\!\eta(y)]\Big\},
372: \end{eqnarray}
373: Notice an essential difference between VM and IP dynamics. In the VM the
374: transition rate is proportional to the inverse degree $k_x$ of the node of
375: the disappearing genotype [Eq.~\eqref{trans::voter}], while in the IP the
376: transition rate is proportional to the inverse degree $k_y$ of the node of
377: the reproducing genotype [Eq.~\eqref{trans::invasion}].
378:
379: For degree-uncorrelated graphs, the transition probabilities are
380: \begin{equation}
381: \label{update::MR-IP}
382: {\bf F}_k(\eta) = \frac{k}{\mu_1} n_k \rho (1-\rho_k)\,,
383: \quad {\bf B}_k(\eta) = \frac{k(1-s)}{\mu_1} n_k (1-\rho) \rho_k\,.
384: \end{equation}
385: Consequently the time evolution of $\rho_k$ is given by, in analogy with
386: Eq.~\eqref{evolution::rhok::MR},
387: $\partial_t \rho_k = \frac{k}{\mu_1}[\rho-\rho_k + s \rho_k(1-\rho)]$,
388: %\begin{equation}
389: %\label{rhok-dot}
390: %\partial_t \rho_k = \frac{k}{\mu_1}[\rho-\rho_k + s \rho_k(1-\rho)]\,,
391: %\end{equation}
392: from which low-order moments obey the equations of motion:
393: \begin{eqnarray*}
394: \partial_t \omega_{-1} &=& \frac{s}{\mu_1 \mu_{-1}} \rho(1-\rho) \,,\\
395: \partial_t \rho &=& \rho - \omega + s \omega(1-\rho)\,, \\
396: \partial_t \omega &=& \frac{\mu_2}{\mu_1}[\rho-\omega_2 + s \omega_2(1-\rho)]\,.
397: \end{eqnarray*}
398: In contrast to the VM, the conserved quantity in the unbiased IP is
399: $\omega_{-1}$, the inverse degree-weighted frequency. For the biased IP,
400: $\omega_{-1}$ becomes the most slowly changing quantity (see
401: Fig.~\ref{traj}). Hence we transform all derivatives with respect to
402: $\rho_k$ in the generator to derivatives with respect to $\omega_{-1}$ to
403: yield
404: \begin{equation}
405: \label{diffusion::invasion}
406: G = \frac{\omega_{-1}(1-\omega_{-1})}{\mu_1 \mu_{-1}} \left[
407: s\frac{\partial}{\partial \omega_{-1}} + \frac{1}{N}(1-\frac{s}{2})\frac{\partial^2}{\partial \omega_{-1}^2}\right]\,,
408: \end{equation}
409: from which, in close analogy with our previous analysis of the VM, the
410: fixation probability is
411: \begin{equation}
412: \label{fixation::invasion}
413: \Phi(\omega_{-1}) = \frac{1-e^{-sN\omega_{-1}/(1-s/2)}}{1-e^{-sN/(1-s/2)}}.
414: \end{equation}
415:
416: From Eq.~\eqref{fixation::invasion}, the effective population size $N_{\rm
417: eff}$ equals $N$, contrary to VM dynamics (Eq.~\eqref{fixation::voter}).
418: More strikingly, the fixation probability of a single mutant acquires the
419: non-trivial dependence of the degree $k$ of the occupied node
420: (Fig.~\ref{surv-one})
421: \begin{equation}
422: \label{fixation::invasion-k}
423: \Phi_1=
424: \begin{cases}
425: 1/(Nk\mu_{-1}) & s\ll 1/N \,; \\
426: s/(k\mu_{-1}) & 1/N \ll s \ll 1\,.
427: \end{cases}
428: \end{equation}
429:
430: To conclude, mutants are more likely to fixate in the voter model (VM) when
431: they are initially on high-degree nodes [Eq.~\eqref{fixation::voter-k}],
432: while in the invasion process (IP) fixation is more probable when mutants
433: start on low-degree nodes [Eq.~\eqref{fixation::invasion-k}]. This behavior
434: is understandable simply. In the VM, a well-connected individual is more
435: likely to be asked his opinion before he asks one of his neighbors. In the
436: IP, a mutant on a high-degree node is more likely to be invaded by a neighbor
437: before the mutant itself can invade. Thus network heterogeneity leads to
438: effective evolutionary heterogeneity.
439:
440: We can also understand the evolution when a mutant appears at a random node
441: on a graph. In the selection-dominated regime ($sN_{\rm eff} \gg 1$) of the
442: VM, we average Eq.~\eqref{fixation::voter-k} over all nodes and find that the
443: fixation probability on degree-uncorrelated graphs is smaller by a factor
444: $\mu_1^2/\mu_2\le1$ than that on regular graphs. Thus a heterogeneous graph
445: is an inhospitable environment for a mutant that evolves by VM dynamics.
446: Conversely, performing the same average of Eq.~\eqref{fixation::invasion-k}
447: over all nodes, the fixation probability for the IP is the same on all
448: degree-uncorrelated graphs. Finally, in the small-selection limit ($sN_{\rm
449: e} \ll 1$), the node average fixation probability is the same for both the
450: VM and IP on degree-uncorrelated graphs.
451:
452: \acknowledgments{We gratefully acknowledge financial support from the Swiss
453: National Science Foundation under Grant No.\ 8220-067591NSF (TA) as well as
454: US National Science Foundation Grant No.\ DMR0535503 (SR and VS). }
455:
456: \begin{thebibliography}{99}
457:
458: \bibitem{epidemics} R. M. Anderson and R. M. May {\it Infectious Diseases in
459: Humans} (Oxford University Press, Oxford, 1992); R. Pastor-Satorras and
460: A. Vespignani, Phys.\ Rev.\ Lett.\ {\bf 86}, 3200 (2001); M. Barth\'elemy,
461: A. Barrat, R. Pastor-Satorras, and A. Vespignani, Phys.\ Rev.\ Lett.\ {\bf
462: 92}, 178701 (2004).
463:
464: \bibitem{fads} D. J. Watts, Proc.\ Natl.\ Acad.\ Sci.\ USA {\bf 99}, 5766
465: (2002); A. Gr\"onlund and P. Holme, Adv.\ in Complex Syst.\ {\bf 8}, 261
466: (2005); J. Bendor, B. A. Huberman, and F. Wu, arXiv.org:physics/0509217.
467:
468: \bibitem{coop} F. C. Santos and J. M. Pacheco, PRL {\bf 95}, 098104 (2005).
469:
470: \bibitem{moran::MP} P. A. P. Moran, Proc.\ Camb.\ Phil.\ Soc.\ {\bf 54}, 60
471: (1958).
472:
473: \bibitem{invasion} C. Taylor, D. Fudenberg, A. Sasaki, and M. A. Nowak,
474: Bull.\ Math.\ Biol.\ {\bf 66}, 1621 (2004); A. Traulsen, J. C. Claussen, and
475: C. Hauert, Phys.\ Rev.\ Lett.\ {\bf 95}, 238701 (2005); T. Antal and I.
476: Scheuring, arXiv:q-bio.PE/0509008.
477:
478: \bibitem{lieberman::EDG} E. Lieberman, C. Hauert, and M. A. Nowak, Nature
479: {\bf 433}, 312 (2005).
480:
481: \bibitem{kimura::NTE} M. Kimura, {\it The Neutral Theory of Molecular
482: Evolution}, (Cambridge University Press, Cambridge UK, 1983).
483:
484: \bibitem{ewens::MPG} W. J. Ewens, {\it Mathematical Population Genetics}
485: (Springer, New York, 2004).
486:
487: \bibitem{liggett} T. M. Liggett {\it Interacting Particle Systems},
488: (Springer-Verlag, New York, 2004); {\it Stochastic Interacting Systems:
489: Contact, Voter, and Exclusion Processes}, (Springer-Verlag, New York,
490: 1999).
491:
492: \bibitem{krapivsky::VM} P. L. Krapivsky, Phys.\ Rev.\ A {\bf 45}, 1067
493: (1992); L. Frachebourg and P. L. Krapivsky, Phys.\ Rev.\ E {\bf 53}, R3009
494: (1996).
495:
496: \bibitem{sood::VM} V. Sood and S. Redner, Phys.\ Rev.\ Lett.\ {\bf 94}, 178701 (2005).
497:
498: \bibitem{SEM} K. Suchecki, V. M. Eguiluz, M. San Miguel, Europhys.\ Lett.\
499: {\bf 69}, 228 (2005).
500:
501: \bibitem{C05} C. Castellano, AIP Conf.\ Proc.\ {\bf 779}, 114 (2005).
502:
503: \bibitem{karlin} N. G. van Kampen, {\it Stochastic Processes in Physics and
504: Chemistry}, 2nd ed.\ (North-Holland, Amsterdam, 1997); S. Redner, {\it A
505: Guide to First-Passage Processes}, (Cambridge University Press, New York,
506: 2001).
507:
508: \bibitem{barton::1997} M. C. Whitlock and N. H. Barton, Genetics {\bf 146}, 427 (1997).
509:
510: \end{thebibliography}
511: \end{document}
512: