Is a fixed-point equation always nonlinear?

No. Affine mappings produce linear equations after rearrangement, as in the worked two-variable example.

Can a root-finding method solve one?

Yes. It can solve the residual equation T(x) − x = 0, subject to the method's requirements.

Does uniqueness imply that direct iteration converges?

No. The location and number of solutions do not alone determine the behavior of repeated application.

AI Glossary / Mathematics and optimization / Intermediate

Fixed-point equation

Definition

A fixed-point equation is an equation of the form x = T(x), where T maps an allowed space into itself. A solution is a value left unchanged by the mapping. Such equations can describe equilibria, optimization conditions or consistency relationships between quantities. Writing a problem in this form specifies what a solution must satisfy; it does not by itself specify a reliable numerical method for obtaining that solution.

What it means

The unknown appears both as the result and as the input to a rule. This is common when a quantity depends on consequences of its own value or on other quantities updated together. The unknown can be a scalar or a vector, so the equation can represent several coupled relationships. A fixed-point equation differs from a fixed-point iteration: the equation defines the target, while iteration repeatedly applies a mapping in an attempt to reach that target. Keeping the two separate prevents misleading convergence claims.

How it works

Start with a mapping T and its domain. Solutions satisfy the residual equation R(x) = T(x) − x = 0. Conversely, a root equation F(x) = 0 can often be written as x = x − ηF(x) for a nonzero scalar η, provided the mapping is defined where needed. Different choices of η or different algebraic rearrangements preserve the roots but can change iteration behavior. If a rearrangement divides by an expression that can vanish, its domain must be checked because it may discard solutions. Domain restrictions are part of the equation, not a minor implementation detail.

Three practical examples

A pair of coupled calibration relationships can be written as one vector fixed-point equation for two unknown settings.
An optimization team can express a zero-gradient condition through a gradient-step operator to connect minimization with fixed-point methods.
A numerical analyst can transform a fixed-point equation into a root problem and compare suitable solvers while verifying the same original residual.

Think of it this way

Think of two departments agreeing on plans that must remain consistent when each department recalculates its requirements from the other's plan. A mutually consistent pair is unchanged by the specified recalculation rule. The analogy helps explain simultaneous equations, but real negotiation may change the rules themselves. A mathematical fixed-point equation assumes a defined mapping and a stated domain during the analysis.

Worked example

Consider the coupled equations u = (v + 1)/2 and v = (u + 2)/3. Define T(u, v) = ((v + 1)/2, (u + 2)/3). Substituting v = (u + 2)/3 into the first equation gives 2u = (u + 2)/3 + 1. Multiplying by 3 yields 6u = u + 5, so u = 1 and then v = 1. Direct substitution verifies T(1, 1) = (1, 1). At the candidate (0.9, 1.1), T gives (1.05, 0.966666...), so the residual is approximately (0.15, −0.133333...). This candidate is not an exact solution even though both coordinates are close to 1. The equivalent linear system is 2u − v = 1 and −u + 3v = 2, with determinant 5, confirming a unique solution. For a separate domain caution, the equation x = x² has solutions 0 and 1. Dividing both sides by x produces 1 = x only for x ≠ 0, losing the solution 0. An algebraic rewrite can therefore fail to preserve the solution set if its required restriction is forgotten.

Illustrative case study

In a fictional planning prototype, Aspen Services models two interdependent resource settings with simultaneous equations. The analyst implements a proposed update and separately codes the original residual equations. A candidate that appears stable under rounded display values still has an unacceptable residual in one component, so the team continues the numerical investigation. They also confirm the allowed range for each setting. This illustrates equation checking and domain discipline, not a claim that the model captures every operational constraint.

Common misconceptions and mistakes

Writing x = T(x) does not make T a contraction. The equation can have no solution, one solution or many. Repeated updates can diverge even when the equation has a unique solution. A root of T itself is generally different from a root of T(x) − x. Squaring, dividing or applying an inverse during rearrangement can introduce or remove candidates unless the domain is handled carefully. In vector problems, checking only one component can miss inconsistency elsewhere. Finally, a numerical solver's return value should be tested in the original equation rather than trusted because it has plausible units or magnitude.

Business use

For a model built from coupled rules, document each variable's units and allowed values before solving. Preserve the original equations alongside any transformed representation. Check candidate solutions against both residual tolerances and domain restrictions. If the model represents a business equilibrium, evaluate whether the equilibrium is useful and whether the assumptions remain credible. Choose a solver based on the mapping's structure and available derivatives or bounds. A change in an equation or scaling convention can affect numerical behavior, so include those details in reproducible model records.

Check your understanding

Why is dividing x = x² by x an incomplete solution procedure over all real numbers?

Answer: Division assumes x is nonzero. It excludes x = 0, which satisfies the original equation, so that case must be checked separately.

Related terms

Fixed point
Root finding
Fixed-point iteration
Domain
residual equation
simultaneous equations

Learn next

solve the worked two-variable equations by substitution and by a linear-system method, then check both answers in T. Start a direct iteration from (0, 0) and compare its intermediate residuals. Identify a different rearrangement of the same equations and examine whether its iteration behaves similarly.

Sources

Sources: Jung and colleagues, The Aalto Dictionary of Machine Learning, fixed-point equation, https://aaltodictionaryofml.github.io/ADictML_Main.pdf ; SciPy, Optimization and root finding reference, https://docs.scipy.org/doc/scipy/reference/optimize.html ; Parikh and Boyd, Proximal Algorithms, https://stanford.edu/~boyd/papers/pdf/prox_algs.pdf . The coupled equations, residual calculation and domain counterexample are original instructional examples.

Previous: Fixed-point computation Next: Fixed-point iteration