Finite-Dimensional State Evolution Recursion

• Finite-dimensional state evolution recursion is defined as a framework where a system’s state evolves recursively via finite-dimensional recurrence relations, as seen in matrix product states and iterative algorithms.
• The approach employs methodologies including spectral analysis, AMP algorithm state tracking, and recursion operators to rigorously derive structural properties and provide error bounds.
• Applications span interacting particle systems, Hamiltonian dynamics, and renewal processes, enabling precise modeling, prediction, and control across various scientific fields.

Finite-dimensional state evolution recursion refers to mathematical frameworks wherein the evolution of a system’s state is governed by a set of recurrence relations or a recursive operator acting on a finite-dimensional space. Research on this topic spans statistical mechanics, stochastic processes, evolution equations, statistical learning, algorithm analysis, and control theory. The following sections detail key methodologies, structural properties, and applications from canonical works in the literature.

1. Matrix Product Stationary States and Rank Constraints

In one-dimensional stochastic models such as the asymmetric simple exclusion process (ASEP), finite-dimensional state evolution is realized via matrix product stationary states (MPSS) (Hieida et al., 2011). The stationary state vector $P_L$ is written in matrix product form: $$P_L = \frac{1}{Z_L} \langle W | (E D)^{\otimes L} | V \rangle,\qquad Z_L = \langle W | (E + D)^L | V\rangle$$ where $E$ and $D$ are $M\times M$ matrices, $M$ finite, and $\langle W|$, $|V\rangle$ are boundary vectors. To verify the existence of a finite-dimensional representation, the system’s stationary state is arranged into a matrix $P^{(m,n)}$ for small $L$ and the rank of $$P_L = \frac{1}{Z_L} \langle W | (E D)^{\otimes L} | V \rangle,\qquad Z_L = \langle W | (E + D)^L | V\rangle$$0 is analyzed. A necessary condition for $$P_L = \frac{1}{Z_L} \langle W | (E D)^{\otimes L} | V \rangle,\qquad Z_L = \langle W | (E + D)^L | V\rangle$$1-dimensional MPSS is $$P_L = \frac{1}{Z_L} \langle W | (E D)^{\otimes L} | V \rangle,\qquad Z_L = \langle W | (E + D)^L | V\rangle$$2. For ASEP, two-dimensional MPSSs exist exactly when $$P_L = \frac{1}{Z_L} \langle W | (E D)^{\otimes L} | V \rangle,\qquad Z_L = \langle W | (E + D)^L | V\rangle$$3

By reverse-engineering finite-dimensional matrices $$P_L = \frac{1}{Z_L} \langle W | (E D)^{\otimes L} | V \rangle,\qquad Z_L = \langle W | (E + D)^L | V\rangle$$4 and $$P_L = \frac{1}{Z_L} \langle W | (E D)^{\otimes L} | V \rangle,\qquad Z_L = \langle W | (E + D)^L | V\rangle$$5 using linear equations based on matrix forms for small $$P_L = \frac{1}{Z_L} \langle W | (E D)^{\otimes L} | V \rangle,\qquad Z_L = \langle W | (E + D)^L | V\rangle$$6, a recursion for the stationary state at larger $$P_L = \frac{1}{Z_L} \langle W | (E D)^{\otimes L} | V \rangle,\qquad Z_L = \langle W | (E + D)^L | V\rangle$$7 arises. Similarity transformations simplify the computation by normalizing basis vectors via $$P_L = \frac{1}{Z_L} \langle W | (E D)^{\otimes L} | V \rangle,\qquad Z_L = \langle W | (E + D)^L | V\rangle$$8. This construction allows the stationary state to be generated recursively for any $$P_L = \frac{1}{Z_L} \langle W | (E D)^{\otimes L} | V \rangle,\qquad Z_L = \langle W | (E + D)^L | V\rangle$$9.

2. State Evolution in High-dimensional Statistical Algorithms

State evolution recursions are central to the asymptotic analysis of iterative algorithms, notably Approximate Message Passing (AMP) and generalized first-order/saddle-point methods (Javanmard et al., 2012, Celentano et al., 25 Jul 2025). For AMP, the state evolution tracks the empirical distribution of iterates through recursive relations on low-dimensional statistics, such as covariance matrices: $E$0 where $E$1 is the algorithm’s nonlinearity, $E$2 is Gaussian with covariance $E$3, and $E$4 are weights. For generalized state evolutions that go beyond separable updates, recursion is realized via the existence and uniqueness of a structured fixed point in an appropriate Hilbert space. Rigorous finite-sample guarantees are proven by lifting iterates to L$E$5 spaces and using sequential Gaussian conditioning and comparison inequalities, establishing uniform concentration of empirical updates around the predicted state evolution for a number of iterations up to $E$6.

3. Recursion Operators in Evolution Equations and Hamiltonian Systems

Finite-dimensional state evolution recursion is abstracted by recursion operators in the context of evolution equations and perturbed Hamiltonian/bi-Hamiltonian systems (Nadjafikhah et al., 2012). If the system is given as $E$7, a linear operator $E$8 is a recursion operator if

$E$9

where $D$0 is the Fréchet derivative of $D$1. For bi-Hamiltonian systems, $D$2 connects two Hamiltonian forms, and repeated application $D$3 generates an infinite hierarchy of approximate symmetries and conservation laws. This recursive structure is analogous to finite-dimensional state evolution in the sense that the state space evolves via iteration of $D$4, with correction terms that scale with the perturbation parameter.

4. Spectral and Fixed-point Analysis in Evolution Equations

Steady-state analysis of evolution equations with finite-dimensional nonlinearities proceeds by decomposing the steady-state problem into a spectral (eigenvalue) part and a nonlinear fixed-point structure (Calsina et al., 2014). For a family of linear operators $D$5 (generators of strongly continuous semigroups),

$D$6

the existence of zero eigenvalue (stationary state) is established, and the solution reduces recursively—a positive eigenvector is coupled with a nonlinear fixed-point in the environmental operator $D$7. For monotone finite-dimensional nonlinearities, fixed-point theorems yield existence of steady states and structure the state evolution by recursive projection onto the simplex of normalized vectors.

5. Discrete Canonical Evolution and Symplectic Recursion

Discrete linear canonical evolution for finite-dimensional systems, as in quadratic action dynamical models, is structured through explicit affine recursion relations (Káninský, 2020): $D$8 where $D$9 and $M\times M$0 are constructed from matrices classified by the system’s quadratic form, and $M\times M$1 parametrize nonuniqueness in irregular evolution steps. Canonical Hamiltonian structure is maintained via symplectic reduction, and introducing adapted coordinates via symplectic transformations simplifies the recursion to a trivial shift over physical degrees of freedom. In models such as massless scalar fields on spacetime lattices, SVD and null-space projectors clarify the propagation and constraints in the evolution.

6. Markov Renewal Chains and Multidimensional Time Recursion

Multi-time Markov renewal chains, a generalization of classical renewal processes, operate on finite-state spaces with multidimensional time (Kordalis et al., 3 Aug 2025). Recursion enters through convolutions of matrix-valued sequences over $M\times M$2: $M\times M$3 Convolutional inverses are computed via a Neumann-type series and adapted Gauss–Jordan algorithms for multidimensional matrix sequences, with FFT acceleration. The evolution of system state is described by a renewal equation, solved recursively using the convolutional inverse,

$M\times M$4

encompassing the full Markov renewal structure over multiple time components.

7. Recursive Representations for Jump-diffusion SDEs

In multivariate SDEs with time–state-dependent jumps, the jump component can be “peeled away” via a recursive approximation using jump times as information relay points (Qiu et al., 2021). Approximate solutions $M\times M$5 are defined recursively: $M\times M$6 At each step, the process is simulated up to the $M\times M$7-th jump, and the next iterate is conditioned on the state at that stopping time. Decoupling is achieved analytically via change of measure and explicit discounting (using $M\times M$8 factors), so that the recursion mimics a Picard iteration in PDE theory. The convergence rate is exponential in the number of jumps considered $M\times M$9, providing systematic error bounds and facilitating numerical computation for high-dimensional problems.

8. Higher-order Evolution Equations and Operator Recurrences

The connection between first-order and higher-order evolution equations is formalized using operator logarithms and systematic recurrence formulas (Iwata, 2022): $M$0 with base case $M$1 from the first-order evolution. In finite-dimensional Banach spaces (e.g., $M$2), higher-order dynamics are reconstructed recursively from $M$3, with operator logarithms and factorized forms tying together functional representations analogous to Cole–Hopf and Miura transforms.

9. Applications and Implications

The deployment of finite-dimensional state evolution recursions appears in:

• Exact construction of stationary distributions in interacting particle systems (Hieida et al., 2011),
• Sharp asymptotic and finite-sample analysis of iterative algorithms in high-dimensional optimization and compressed sensing (Javanmard et al., 2012, Celentano et al., 25 Jul 2025),
• Generation of infinite hierarchy of symmetries and conservation laws in Hamiltonian PDEs (Nadjafikhah et al., 2012),
• Recursive solution schemes for nonlocal PIDEs in jump-diffusion settings (Qiu et al., 2021),
• Efficient solution of renewal equations in reliability, seismology, and discrete event systems (Kordalis et al., 3 Aug 2025).

These recursions enable rigorous prediction, control, and simulation in systems where the effective description can be collapsed onto a (potentially low-dimensional) vector or operator recursion, with explicit structural, spectral, or probabilistic bounds underpinning both theory and numerical analysis.