--- title: Instruction Survival Probability (Psi) url: https://www.emergentmind.com/topics/instruction-survival-probability-psi type: topic --- # Instruction Survival Probability (Psi) Searching arXiv for the cited papers to ground the article in current metadata and ensure accurate references. arXiv search query: 1110.0796 Instruction Survival Probability, written here as \(\Psi\), is not a single invariant quantity across the arXiv literature but a family of survival, persistence, return, or non-absorption probabilities indexed by time, scale, or generation. In the surveyed works, \(\Psi\) denotes, depending on context, the probability that a branching system is still alive at time \(n\), that a stochastic process has not yet exited a domain by time \(t\), that a quantum state is still found in its initial configuration, or that an excitation or target has avoided trapping or attenuation. The common structure is probabilistic persistence under dynamics, but the underlying events, asymptotic laws, and analytic tools differ sharply across statistical physics, probability, quantum dynamics, transport theory, and high-energy applications [1110.0796], [2302.08512]. ## 1. Definitions and notational conventions The notation \(\Psi\) is used in the surveyed literature as a generic survival function, but the primary symbols vary by field. In high-dimensional statistical physical models, the survival probability is \(\theta_n=\mathbb P(N_n>0)\), also appearing in OCR-affected versions as \(\Theta_n\) or \(\Omega_n\) [1110.0796]. In exit problems for Lévy processes it is \(\mathbb P^x(\tau_D>t)\), where \(\tau_D\) is the first exit time from a domain \(D\) [1307.0270]. In persistence theory for iterated processes it is \(\Psi(T)=\mathbb P(\sup_{t\in[0,T]} Z_t \le 1)\) [1106.2999]. In quantum dynamics it may be the standard survival probability \(SP(t)=|\langle \Psi(0)|\Psi(t)\rangle|^2\) or the generalized quantity \(SP_q(t)\) [2302.08512]. | Context | Notation in the papers | Definition | |---|---|---| | High-dimensional branching/percolation | \(\theta_n\), \(\Theta_n\), \(\Omega_n\) | \(\mathbb P(N_n>0)\) | | Exit from a domain | \(\mathbb P^x(\tau_D>t)\) | probability of no exit by time \(t\) | | Persistence below a barrier | \(\Psi(T)\), \(p_N(x)\) | probability of staying below a threshold | | Quantum return | \(SP(t)\), \(R(t)\), \(p_s(t)\) | probability of remaining in or returning to the initial state | | Restart/trap models | \(S(t;x_0)\), \(S(t;x_0,q)\), \(P_s(t)\) | probability of no absorption up to time \(t\) | These usages separate into two recurrent meanings. One is **survival against extinction, absorption, exit, or trapping**. The other is **return or occupation fidelity**, where “survival” means persistence of overlap with an initial state. This suggests that \(\Psi\) is best treated as a structural observable rather than a model-specific object. ## 2. Branching, percolation, and critical survival laws In high-dimensional statistical physical models, survival probability is tied to branching structure and super-Brownian scaling. For spread-out lattice trees above \(8\) dimensions, spread-out oriented percolation above \(4+1\) dimensions, and the spread-out contact process above \(4+1\) dimensions, the central result is the universal asymptotic \[ n\,\theta_n \longrightarrow \frac{2}{AV}, \] under convergence of the \(r\)-point functions to those of the canonical measure of super-Brownian motion and a self-repellence bound [1110.0796]. The same framework uses the cluster tail estimate \[ \mathbb{P}(|C(0)|>k)\le C_c k^{-1/2} \] and the self-repellent survival property \[ \mathbb{P}(A_m\to n\mid \mathcal F_m)\le C_e\,\frac{N_m}{n-m}. \] Conditioned on survival up to time \(n\), the rescaled particle counts converge in finite-dimensional distributions to Feller branching diffusion started from an exponential initial distribution with mean \(A^2Vt/2\) [1110.0796]. Branching Brownian motion yields a different class of survival laws. For near-critical branching Brownian motion with drift \(\mu=\sqrt{2-\epsilon}\), killed at the origin, the survival probability \(Q_\mu(x)\) has a nontrivial limit on the critical spatial scale \(L=\pi/\sqrt{\epsilon}\): \[ \lim_{\epsilon\to 0} Q_\mu(L+x)=\theta(x), \] where \(\theta\) is a travelling-wave solution of the Fisher–KPP equation \[ 2\theta''=\theta'-\theta(1-\theta) \] with boundary conditions \(\theta(-\infty)=1\) and \(\theta(+\infty)=0\) [1009.0406]. Well below the threshold \(L\), the sharp asymptotic becomes \[ Q_\mu(x)\sim C\,L\,\exp\!\big(-\mu(L-x)\big)\,\sin\!\left(\frac{\pi x}{L}\right). \] At the exactly critical drift \(-\sqrt 2\), the process dies out almost surely, but survival to a large finite time has a stretched-exponential form. The extinction-time survival probability \(\Psi(t)=\mathbb P(\zeta>t)\) satisfies \[ \mathbb P(\zeta>t)\asymp e^{\sqrt2 x}\sin\!\Big(\frac{\pi x}{c\,t^{1/3}}\Big)t^{1/3}e^{-(3\pi^2 t)^{1/3}}, \] and for fixed \(x\), \[ \mathbb P(\zeta>t)\asymp x e^{\sqrt2 x}e^{-(3\pi^2 t)^{1/3}} \] [1212.3821]. Here survival is neither algebraic nor purely exponential. In subcritical branching processes in random environment, the survival probability may decay exponentially with a heavy-tail correction. For a BPRE with environment variable \(X\) satisfying \[ p_X(x)=x^{-\beta-1}l_0(x)e^{-\rho x}, \qquad \beta>2,\ \rho\in(0,1), \] the asymptotic survival probability is \[ \mathbb P(Z_n>0)\sim C_0\,\rho^{\,n-1}e^{an}\mathbb P(X>an)\sim C_0\,m^{n-1}b_n, \] so the dominant decay is exponential, modified by a polynomial factor inherited from the tail of \(X\) [1307.3963]. The same paper proves a Yaglom-type conditional limit theorem for \(Z_n\mid\{Z_n>0\}\). ## 3. Persistence, exit, and first-passage formulations A large probability-theoretic literature interprets survival as **non-exit** or **one-sided persistence**. For isotropic unimodal Lévy processes, \(\Psi(t,x)=\mathbb P^x(\tau_D>t)\) is controlled by boundary distance through the renewal function \(V\). In \(C^{1,1}\) domains, \[ \frac{1}{C_*}\,\frac{V(\delta_D(x))}{V(\sqrt t)} \le \mathbb P^x(\tau_D>t) \le C_*\,\frac{V(\delta_D(x))}{V(\sqrt t)} \] for \(0t)\,dt. \] For compact random walks in unbounded space, survival is the probability \(S(t)\) that a target has not yet been reached. The long-time asymptotic law is \[ S(t)\sim \frac{S_0}{t^\theta}, \] with \(\theta=1-d_f/d_w\) in the Markovian fractal setting and \(\theta=1-1/d_w\) for the one-dimensional stationary-increment non-Markovian processes considered in that work [1907.03632]. A central result is that the prefactor is tied to the large-volume mean first-passage time: \[ S_0=\frac{\sin(\pi d_f / d_w)}{\pi K}\,\overline T. \] This shifts attention from persistence exponents alone to quantitatively essential amplitudes. Iterated processes give a persistence exponent transfer rule. For \[ Z_t=X_{|Y_t|}\qquad\text{or}\qquad Z_t=X_{Y_t}, \] with \(Y\) continuous and self-similar of index \(H\), if \[ \mathbb P\!\left(\sup_{t\in[0,T]} X_t\le 1\right)\asymp T^{-\theta}, \] then \[ \mathbb P\!\left(\sup_{t\in[0,T]} X(Y_t)\le 1\right)\asymp T^{-\theta H} \] for one-sided \(X\), while for two-sided \(X\) the exponent becomes \(H(\theta^-+\theta^+)\) [1106.2999]. For two-sided iterated Brownian motion this yields \[ \mathbb P\!\left(\sup_{t\in[0,T]} B(W_t)\le 1\right)\asymp T^{-1/2}. \] Autoregressive processes supply a discrete-time persistence theory. For \[ X_n=a_1X_{n-1}+\cdots+a_pX_{n-p}+Y_n, \] the one-sided survival probability is \[ p_N(x)=\mathbb P\!\left(\sup_{n=1,\dots,N}X_n\le x\right). \] For AR(2), the paper identifies a trichotomy: polynomial decay on the critical line \(P\), fast decay in the region \(E\), and convergence to a positive limit in the region \(C\) [1207.3610]. On \(P\setminus\{(2,-1)\}\), \[ p_N=N^{-1/2+o(1)} \quad (a_2<1), \qquad p_N\asymp N^{-1}\quad (a_2=1), \] while for the integrated random walk point \((2,-1)\), \[ p_N(x)\asymp N^{-1/4}. \] ## 4. Quantum return, generalized fidelity, and non-exponential decay In quantum dynamics, survival probability is often a return probability rather than a non-absorption probability. The standard form is \[ SP(t)=|\langle \Psi(0)|\Psi(t)\rangle|^2 =\left|\sum_\alpha |C_\alpha^{(0)}|^2 e^{-iE_\alpha t}\right|^2 =\left|\int \rho(E)e^{-iEt}\,dE\right|^2, \] where \(\rho(E)\) is the local density of states [2302.08512]. The generalized survival probability introduces a \(q\)-weighted fidelity, \[ SP_q(t)=\frac{1}{\mathcal N_q^2}\left|\sum_{\alpha=1}^{\mathcal D}|C_\alpha^{(0)}|^q e^{-iE_\alpha t}\right|^2, \] with \(q=2\) recovering the standard survival probability and \(q=0\) giving the spectral form factor [2302.08512]. In GOE random matrices, the generalized local density of states remains semicircular and \(\sigma_q\) is nearly independent of \(q\), whereas in the disordered Heisenberg spin-\(\tfrac12\) chain the generalized local density of states is Gaussian but narrows with increasing \(q\), and the long-time power-law exponent \(\gamma_q\) decreases with \(q\). The generalized Rosenzweig–Porter ensemble uses survival probability as a phase diagnostic. For a projected wave packet, the return probability is \[ R(t)=\overline{\big|\langle \Psi(0)|e^{-iHt}|\Psi(0)\rangle\big|^2}. \] Its phase-dependent behavior distinguishes ergodic, multifractal, and localized regimes: oscillatory power-law decay in the ergodic phase, finite-size exponential decay \[ R(t)\sim e^{-2\Gamma(N)t} \] with \(\Gamma(N)\propto N^{D-1}\) in the multifractal phase, and saturation to a finite constant in the localized phase [1805.06472]. The Bixon–Jortner model shows that “survival” need not imply monotone decay. For an initially excited state \(s\), the occupation probability is \[ p_s(t)=|x_s(t)|^2. \] The numerics show a non-exponential short-time region, often an intermediate exponential-like regime \[ p_s(t)=e^{-\gamma t}, \] and then repeated repopulation because the system contains only a countable set of coupled states [2309.03365]. With \(n=26\) states, the continuum decay-rate estimate \[ \gamma=\frac{2\pi V^2}{\epsilon} \] is already accurate for several parameter choices, but the long-time behavior remains recurrent rather than irreversible. This directly counters the common simplification that survival decay in quantum systems is generically exponential. ## 5. Restart, traps, networks, and open-system survival Absorbing traps and restart rules generate another major survival-probability class. In the Sisyphus random walk with equal left-step and reset probabilities, the survival probability is \[ S(t;x_0)=\frac{G_t^{(x_0)}}{2^t}, \] where \(G_t^{(x_0)}\) satisfies an \(x_0\)-generalized Fibonacci-like recurrence, and the late-time behavior is \[ S(t;x_0)\sim a(x_0)\left[1-2^{-(x_0+1)}\right]^t \] [2407.13180]. In the biased version, where the walker steps toward the trap with probability \(q>1/2\) and restarts to \(x_0\) with probability \(1-q\), the asymptotic form is \[ S(t;x_0,q)\sim a(x_0,q)\,[B(x_0,q)]^t, \] with \(B(x_0,q)\) determined by \[ B^{x_0+1}-B^{x_0}+q^{x_0}(1-q)=0 \] [2502.06667]. The same paper defines a critical gap \(x_0^{\text{crit}}(q)\) above which restart improves late-time survival relative to an ordinary biased walk. This shows that restart neither uniformly helps nor uniformly harms survival. For an immobile target surrounded by independent CTRW traps in one dimension, the exact identity \[ P_s(t)=\exp\!\left[-2\rho\,E[M(t)]\right] \] reduces the many-trap problem to the expected maximum \(E[M(t)]\) of a single trap [1203.2859]. The resulting long-time law is stretched exponential, \[ P_s(t)\sim a\,\exp[-b\,t^\theta], \] with \(\theta\) determined by the jump-tail and waiting-time exponents. On Erdős–Rényi random networks, the survival probability of a delta-like excitation is \[ SP(t)=\left|\langle \Psi(0)\vert \Psi(t)\rangle\right|^2. \] The short-time law is quadratic, \[ SP(t\ll 1)\approx 1-\sigma_{ini}^2 t^2 \approx 1-\langle k\rangle t^2, \] followed by a power-law regime governed by \(D_2\) or, for the time-averaged quantity \(C(t)\), by \(\widetilde D_2\), then a correlation hole, and finally saturation near the initial-state inverse participation ratio [2603.15682]. The relative depth of the correlation hole approaches the GOE value \(1/3\) around \(\langle k\rangle\approx 10\). Quantum walks on graphs yield a different mechanism for nonzero asymptotic survival. For the Grover walk on a ladder graph with an absorbing sink, the limiting survival probability is the squared norm of the initial state projected onto the dark subspace, \[ S=\|\Pi_{\mathcal K}\psi(0)\|^2+\|\Pi_{\mathcal M}\psi(0)\|^2 \] [2205.13188]. Depending on whether loops are attached at the corners, the survival can increase and converge exponentially or decrease like \(L^{-1}\). The paper identifies a long-path dark state whose normalization scales like \(L^{-1/2}\), producing the \(L^{-1}\) contribution. This rules out the naive expectation that a larger dark subspace must imply larger survival. Open and attenuating systems extend the same language beyond purely stochastic or unitary settings. In charmonium propagation through a hot medium, survival is the modulus squared of a Green-function transition amplitude and receives comparable contributions from Debye screening and absorption [1409.5147]. In leading-neutron production in DIS, the rapidity-gap survival probability \(S^2\) is the exact-to-Born ratio of cross sections, and the computed values are small and decrease with energy [1204.0675]. ## 6. Asymptotic regimes, methods, and recurring misconceptions The surveyed literature does not support a single canonical decay law for \(\Psi\). Instead, it exhibits a taxonomy of asymptotics: algebraic decay \(S(t)\sim S_0/t^\theta\) for compact first-passage problems [1907.03632]; mean-field \(2/n\)-type survival in high-dimensional branching models through \[ n\Psi(n)\to \frac{2}{AV} \] [1110.0796]; stretched-exponential decay \(\exp[-(3\pi^2 t)^{1/3}]\) in critical branching Brownian motion [1212.3821]; ordinary exponential decay in restart walks and finite-size multifractal dynamics [2407.13180], [1805.06472]; and non-monotone decay with revivals in discrete quantum spectra [2309.03365]. This suggests that the asymptotic class of \(\Psi\) is set by geometry, spectral structure, conditioning, and effective state-space size rather than by the survival label itself. Several methodological themes recur. Weak convergence to the canonical measure of super-Brownian motion and \(r\)-point function scaling drive the high-dimensional survival asymptotics [1110.0796]. Barrier constructions based on the renewal function \(V\) produce sharp exit-time and survival bounds for unimodal Lévy processes [1307.0270]. Wiener–Hopf factorization controls long-time survival probabilities and lower-tail behavior for broad classes of Lévy processes [2501.04218]. Fourier transforms of the local density of states organize quantum survival laws [2302.08512], while Pollaczek–Spitzer formulas and subordination control target survival among CTRW traps [1203.2859]. Two misconceptions are repeatedly contradicted. First, exponential decay is not universal: some systems display algebraic, stretched-exponential, or revival-dominated survival laws [1907.03632], [1212.3821], [2309.03365]. Second, structural enlargement does not necessarily increase survival: restart can help or hurt depending on \(x_0\) and \(q\) [2502.06667], and extra dark states can coincide with decreasing asymptotic survival in ladder-graph Grover walks [2205.13188]. Taken together, these results define \(\Psi\) as a unifying but non-uniform observable. It measures persistence under dynamics, yet its precise interpretation ranges from extinction avoidance to fidelity retention, from domain non-exit to gap preservation. Its value as a research object lies precisely in that breadth: the same probabilistic syntax supports mean-field branching limits, persistence exponents, random-matrix diagnostics, restart thresholds, trap-induced stretched exponentials, and attenuation factors in open many-body and high-energy systems. Source: https://www.emergentmind.com/topics/instruction-survival-probability-psi