Maximal Interval of Existence

• Maximal Interval of Existence is the largest time span on which a unique, regular solution exists for an initial value problem in evolution equations.
• It is characterized by the blow-up of solution norms or loss of regularity, which signifies the breakdown of well-posedness in both deterministic and stochastic frameworks.
• This concept is vital for analyzing blow-up phenomena, lifespan estimates, and the applicability of methods across ODEs, PDEs, FBSDEs, and dynamical low-rank approximations.

A maximal interval of existence is, in the context of evolution equations (ODEs, PDEs, FBSDEs, and related frameworks), the largest interval—typically of the form $[0,T^*)$ or $(t_{\min}, T^*)$—on which a unique, regular solution to an initial value problem exists according to the analytic or probabilistic criteria appropriate to the setting. The endpoint $T^*$ is often characterized by loss of regularity, blow-up of derivatives or the solution itself, or by other obstructions to further well-posed extension. This notion is fundamental in determining the temporal domain of meaningful modeling for dynamical systems, the analysis of singularities, and the structure of solution spaces in both deterministic and stochastic frameworks.

1. Classical ODE and PDE Frameworks

In the classical theory of ODEs, specifically the Cauchy–Lipschitz (Picard–Lindelöf) setting, the maximal interval of existence $[0, T^*(x))$ for the solution $X(t, x)$ of $\dot{X}(t) = b(t, X(t))$, $X(0) = x$, is defined as the supremum of $t$ for which the trajectory remains in the domain $\Omega$ and the solution remains unique and differentiable. If $T^*(x)<T$, then the solution either escapes every compact subset of $\Omega$ (explosion to infinity), or hits the boundary $\partial\Omega$ in finite time.

For nonlinear PDEs, such as the nonlinear heat equation $\partial_t u - \Delta u = |u|^\alpha u$ or semilinear wave equations $\Box u = |u|^p$, the maximal existence time $T^*(u_0)$ or $T(\varepsilon)$ is defined as the supremum over all $T$ such that a (classical, mild, or weak) solution exists on $[0,T)$, with initial data $u_0$ or specified small-data scaling parameter $\varepsilon$ (Tayachi et al., 2022, Zhou et al., 2011, Takamura et al., 2011).

In these contexts, the endpoint $T^*$ generally coincides with the first time a norm of $u$ (such as $L^\infty$ or $H^1$) blows up, or the first time at which regularity is lost. The quantitative estimates of $T^*$ as a function of parameters in initial data or nonlinear exponents are a major area of analysis, leading to explicit lower and upper bounds (lifespan estimates) as a function of data magnitude and regularity.

2. Maximal Intervals in Non-Smooth and Measure-Driven Dynamics

The theory has been generalized to non-smooth vector fields, as developed in the DiPerna–Lions and Ambrosio–Colombo–Figalli frameworks (Ambrosio et al., 2014). In this setting, for vector fields $b$ only locally integrable or in $W^{1,p}_{\mathrm{loc}}$, the notion of a maximal regular flow is introduced. The associated explosion time $T_\Omega(x)$ denotes the supremum over all times such that the almost-everywhere defined flow $X(t, x)$ remains inside the domain, with analogous pushforward and blow-up characterizations.

Proper blow-up—i.e., the solution tending to infinity in configuration or via a suitable Lyapunov function $V(X(t, x))\to\infty$ as $t \to T_\Omega(x)^-$—may require global conditions, such as bounded compression or global control on $\operatorname{div}\,b$. In the absence of such, only weaker assertions (limsup blow-up) can be made, as illustrated by counterexamples with recurrent trajectories that do not escape all compacts before explosion.

3. Stochastic and Backward-Forward Systems

In the analysis of forward-backward stochastic differential equations (FBSDEs), the maximal interval of existence, denoted $I_{\max}$, for a decoupling field is the union of all time intervals $[t, T]$ on which well-posed variational solutions exist (Fromm et al., 2013). The form of $I_{\max}$ is either $[0, T]$ (global in time) or $(t_{\min}, T]$, where the left endpoint $t_{\min}$ is determined by loss of regularity, typically explosion of the derivative $\|\nabla_x u(t, \cdot)\|_\infty$, as exemplified in explicit solvable cases.

Continuation criteria, based on a priori Lyapunov-type estimates on such derivatives, can in favorable cases guarantee $I_{\max}=[0, T]$. If the continuation criterion is violated (e.g., the critical value in the Lipschitz constant product is approached), the solution cannot be extended further leftward in time.

4. Maximal Interval in Dynamical Low-Rank Approximation

For dynamical low-rank (DLR) approximations of random semi-linear evolution equations, solutions are curves in a smooth manifold $M_S$ of rank-$S$ fields; maximal existence is the largest temporal interval $[0, t_{\max})$ for which a unique DLR solution exists (Kazashi et al., 2020). The obstruction to continuation (blow-up alternative) is the loss of invertibility of the spatial Gram matrix, i.e., when the solution's spatial basis ceases to be linearly independent, which signals that the rank-$S$ manifold is no longer "adequate" to resolve the dynamics.

The continuation (or maximal interval) is thus given by the time before $\|Z_U^{-1}(t)\|$ becomes unbounded. At this point, one may project orthogonally to a lower-rank manifold, leading to extension with reduced representational complexity.

5. Explicit Lifespan Estimates and Asymptotics

Significant research is devoted to explicit estimates and asymptotics of the maximal interval $T^*(\lambda)$ or $T(\varepsilon)$ as a function of data magnitude and structural parameters, particularly in critical regimes where nonlinearity and data scaling are delicately balanced. For the critical semilinear wave equation in four dimensions, sharp asymptotics $\log T(\varepsilon) \sim \varepsilon^{-2}$ have been established for both upper and lower bounds, closing the optimality gap in the Strauss conjecture for lifespan estimates (Takamura et al., 2011, Zhou et al., 2011). Analogous scaling formulas arise in the nonlinear heat equation, Hardy–Hénon parabolic equations, and other nonlinear evolution settings (Tayachi et al., 2022).

These estimates not only reveal the dependence of the maximal interval on norms and decay/singularity properties of initial data, but also guide the qualitative understanding of blow-up formation and propagation.

6. Methodological Principles and Continuation Criteria

Across analytic, probabilistic, and abstract evolution settings, maximal intervals of existence are defined by:

• Local well-posedness guaranteed by contraction, fixed-point, or semigroup arguments;
• Extension via concatenation or bootstrapping, provided analyticity or regularity bounds are maintained;
• Characterization of the limiting time $T^*$ via blow-up of norm, loss of invertibility (in low-rank methods), explosion of derivatives (in FBSDE decoupling fields), or approach to the boundary (in ODE theory);
• Deduction of continuation or obstruction by a priori estimates, barrier arguments, and, where possible, explicit calculation of critical exponents or rates.

The following table summarizes the maximal interval concepts in selected frameworks:

Setting	Maximal Interval Notation	Continuation/Obstruction Mechanism
Classical ODE (Cauchy–Lipschitz)	$[0, T^*(x))$	Trajectory exits domain or diverges
Semilinear/parabolic PDE	$[0,T^*)$ or $[0, T(\varepsilon))$	Blow-up in solution norm or derivative
FBSDE (decoupling field)	$I_{\max}$	Explosion of $\|\nabla_x u\|$, derivative blow-up
DLR approximation	$[0, t_{\max})$	Loss of spatial basis rank (Gram matrix singularity)
Non-smooth flows (DiPerna–Lions/ACF)	$[0, T_\Omega(x))$	Weak or proper blow-up of Lyapunov function $V$

7. Applications and Illustrative Examples

Applications of the maximal interval of existence span deterministic dynamics, stochastic processes, numerical and model-reduction methods, and the rigorous analysis of critical phenomena in nonlinear PDEs. Notable examples include:

• Lifespan of solutions to critical wave and heat equations under scaling of small/large initial data (Takamura et al., 2011, Tayachi et al., 2022);
• Maximal regular flows for non-smooth vector fields, which generalize deterministic ODE solution theory to the setting of measures and weak/renormalized solutions (Ambrosio et al., 2014);
• Temporal validity of dynamical low-rank methods in uncertainty quantification and random evolution (Kazashi et al., 2020);
• Backward-forward SDE solution domains in stochastic control and finance, with explicit derivative explosion as the mechanism cutting off existence (Fromm et al., 2013).

These frameworks collectively illustrate the centrality of maximal intervals across modern analysis and stochastic modeling, encapsulating the delicate interplay between regularity, data, nonlinearity, and the onset of singularities or degeneration.