Fixed-Point Formulation (FPF)

• Fixed-Point Formulation (FPF) is a mathematical method that defines solutions as fixed points of an operator satisfying x = T(x).
• It is applied in quantum mechanics, signal recovery, power systems, and data science to develop robust and convergent algorithms.
• FPF leverages iterative schemes and operator theory to ensure scalability and efficiency across diverse computational and engineering problems.

A fixed-point formulation (FPF) is a mathematical and algorithmic paradigm in which the solution to a problem is characterized as a fixed point of a suitably constructed operator or constraint. In FPFs, the object of interest—be it a field, vector, function, or quantum state—must satisfy the relation $x = T(x)$ for a given mapping $T$. This unifying perspective underpins diverse methodologies in quantum theory, nonlinear signal recovery, power systems analysis, optimization, variational PDEs, data science, and more, enabling the systematic development of robust and convergent algorithms.

1. Fundamental Concepts and Mathematical Structure

The core principle of FPF is to recast the original problem—potentially involving equations, inclusions, feasibility, boundary or event constraints—as the search for a solution that is invariant under an operator. Typically, one constructs $T$ such that the set of its fixed points coincides with the set of solutions to the original problem, i.e., $x^* = T(x^*)$. This approach leverages powerful existence, uniqueness, and convergence results for Banach contractions, nonexpansive mappings, averaged operators, monotone inclusions, and other operator-theoretic constructs (Combettes et al., 2020).

FPFs appear in both linear and nonlinear settings; $T$ may encode gradient-descent, a projection onto constraints, a block-response map in games, or the imposition of measurement events on quantum histories. The iterative solution of fixed-point equations yields a range of classical and modern algorithms, including Picard iteration, Krasnosel’skiĭ–Mann schemes, Douglas–Rachford and forward–backward splitting, block-iterative projections, and data-driven operator learning (Combettes et al., 2020, Heaton et al., 2021).

2. Applications Across Domains

FPFs provide a modeling and computational scaffold for a wide array of fields:

• Quantum Mechanics: The fixed-point formulation articulates quantum events as fixed-point constraints on the universal wavefunction along a Keldysh time contour. Each event corresponds to a requirement that forward and backward branches of the wavefunction coincide in the Hilbert space at a given time, producing an “all-at-once,” time- and event-symmetric block-universe ontology that resolves measurement and retrocausal paradoxes in quantum theory (Ridley et al., 2023).
• Signal and Image Recovery: In nonlinear and inverse problems, measurements and a priori constraints are encoded by constructing a family of (firmly) nonexpansive operators whose common fixed point is the solution. Efficient and convergent block-iterative algorithms can then be deployed, even in the presence of nonlinear transformations or inconsistent data (Combettes et al., 2020).
• Power Systems Analysis: Fixed-point reformulations of power-flow equations yield robust, globally convergent algorithms for AC networks. By recasting nodal equations as fixed-point maps, or by geometric intersection of circles encoding bus constraints, it is possible to design linearly convergent iterations that are competitive with or superior to Newton–Raphson, especially in ill-conditioned, large-scale, or heavily loaded regimes (Duque et al., 2024, Guddanti et al., 2018).
• Data Science and Optimization: A vast class of optimization, feasibility, and equilibrium problems are recast as fixed-point problems, enabling the use of unifying algorithmic frameworks and convergence theory (Combettes et al., 2020).

3. Operator Types and Convergence Mechanisms

A variety of operator types are central to FPF theory and algorithm design:

• Banach Contractions: Operators $T$ such that $\|T(x)-T(y)\| \leq \delta\|x-y\|$ with $\delta<1$ yield unique fixed points and linear convergence by Picard iteration.
• Firmly Nonexpansive/Averaged Operators: If $T=(1-\alpha)\Id+\alpha R$ with $R$ nonexpansive and $\alpha\in(0,1)$, then convergence to a fixed point can be ensured, typically via the Krasnosel’skiĭ–Mann iteration.
• Monotone Operators and Resolvents: For a maximally monotone operator $A$, its resolvent $J_A=(\Id+A)^{-1}$ is firmly nonexpansive and is a cornerstone of splitting algorithms.
• Nonexpansive Operators in Data-Driven Contexts: In learned fixed-point networks, nonexpansiveness is enforced via constraints on neural network regularizers to ensure theoretical convergence (Heaton et al., 2021).

Convergence proofs are established under contractivity, nonexpansiveness, or monotonicity assumptions. Classical results (Banach’s Theorem, Opial’s Lemma, Minty’s Theorem, etc.) guarantee existence, uniqueness, and often strong convergence in finite-dimensional Hilbert spaces or under suitable operator-theoretic conditions (Combettes et al., 2020, Combettes et al., 2020).

4. Representative Methodologies

FPF-inspired methodologies stratify into several major classes, as outlined below:

Problem Type	Operator $T$ Construction	Representative Algorithms
Convex Optimization	Proximal point, gradient,	Picard, Proximal, Forward-Backward
Feasibility	Projectors or reflections	POCS, block-iterative projection
Monotone Inclusions	Splitting resolvents	Douglas–Rachford, Primal–Dual
Power-Flow Analysis	Circle-intersection or	Iterative fixed-point, tensor algorithms
Quantum Histories	Keldysh contour + projections	All-at-once event constraints
Learning/Inverse	Nonexpansive neural maps +	Learned fixed-point networks
	data projections

In quantum mechanics, events as fixed-point constraints produce a tensor product, multi-event structure along a time contour, unifying various conditional rules (Born, ABL) as special cases (Ridley et al., 2023). In AC power-flow, advanced formulations exploit the geometry of intersection in state-space (circle methods) and simultaneous tensor-based solutions to hundreds of thousands of scenarios (Duque et al., 2024, Guddanti et al., 2018).

5. Computational and Algorithmic Properties

FPF-based algorithms are typically simple in structure but display high efficiency, scalability, and robustness subject to the operator chosen and contraction domain:

• Power System FPF: Tensorized implementations execute bulk batched matrix–vector operations with CPU/GPU acceleration, achieving order-of-magnitude speedup relative to NR methods (e.g., ×164 in year-scale time series) (Duque et al., 2024). Circle-intersection FPF achieves O(n) per-iteration cost.
• Block Iterative Methods: In nonlinear inverse problems, block-iterative FPFs maintain convergence by careful activation of constraint operators and relaxation parameters, with Fejér monotonicity ensuring progress even in over-relaxed or inconsistent regimes (Combettes et al., 2020).
• Quantum FPF: The formulation is deterministic, globally consistent, and exhibits atemporal retrocausality, resolving standard paradoxes without requiring explicit backward-in-time dynamics (Ridley et al., 2023).
• Stochastic and Randomized FPFs: Algorithms admitting stochastic operator selection, error perturbations, or coordinate sampling retain convergence guarantees under mild statistical or block-coverage conditions (Combettes et al., 2020).

6. Theoretical and Practical Implications

By adopting the FPF lens, diverse problems in analysis, simulation, and data science can be unified under common algorithmic and convergence theorems. This facilitates not only design of new methodological advances (e.g., plug-and-play regularization using nonexpansive learned denoisers, full signal/image recovery from nonlinear transformations) but also enhances interpretability and global solution properties.

In quantum theory, the FPF achieves true time- and event-symmetry and encodes measurement/disturbance as sharp, non-dynamical constraints, providing a mathematically unified account of histories and their statistics (Ridley et al., 2023). A plausible implication is that atemporal “block-universe” quantum models—via FPF—can systematically evade known causal inconsistency objections affecting transactional and time-symmetric approaches.

In power engineering, FPF guarantees selection of physically meaningful, high-voltage branches by virtue of contraction and impedance ratios, even in the presence of heavy nonlinearity or system ill-conditioning (Duque et al., 2024, Guddanti et al., 2018). In convex and composite optimization, it provides a transparent mechanism to encode splitting, block activation, and stochasticity, and to extend to non-Euclidean settings.

7. Limitations, Extensions, and Outlook

FPF-based methods inherit the strengths and weaknesses of the operator-theoretic machinery. Contraction domain may be non-global (e.g., fixed-point power flow converges only to the high-voltage branch), and for strongly nonlinear or non-monotone systems, convergence may be slow or require continuation/regularization techniques (0806.3514). In high-dimensional or data-driven scenarios, practical performance depends on careful operator design, appropriate relaxation, and hardware utilization.

Recent work emphasizes robust, scalable implementation (e.g., Cholesky-based FPF for state estimation in dynamical systems avoids loss of positive-definiteness and large memory costs), as well as hybridization with learning systems (e.g., feasibility-based fixed-point networks) (Krämer, 2024, Heaton et al., 2021).

A plausible implication is that continued expansion of FPF machinery—especially its fusion with learning, non-Euclidean metrics, and stochastic methods—will further integrate algorithmic, physical, and data-theoretic perspectives across mathematical sciences and engineering.

---

Key References:

• (Ridley et al., 2023) (Quantum: symmetry, event constraints), (Duque et al., 2024) (Power: tensor FPF), (Guddanti et al., 2018) (Power: circle FPF), (Combettes et al., 2020) (Data science: general FPF), (Combettes et al., 2020) (Nonlinear inverse: block FPF), (Heaton et al., 2021) (Learning: feasibility fixed-point networks), (0806.3514) (PDEs: FPF vs NR), (Krämer, 2024) (State-space smoothing: Cholesky FPF).