# Contraction mapping

A **contraction mapping** is a self-map \( f \) of a metric space \( (X, d) \) that shrinks distances by a fixed factor: there is a constant \( q \) with \( 0 \le q < 1 \) such that

\[ d(f(x), f(y)) \le q \, d(x, y) \quad \text{for all } x, y \in X. \]

The constant \( q \) is called the contraction factor or Lipschitz modulus. The central result about such maps is the Banach fixed point theorem, also called the contraction mapping theorem or Banach contraction principle: a contraction on a complete metric space has exactly one fixed point, a point \( x_* \) with \( f(x_*) = x_* \), and it can be found by iterating \( f \) from any starting point.<sup>[1](https://www.ssc.wisc.edu/~jkennan/teaching/ContractionMappingTheorem.pdf)</sup>

| Key fact | Detail |
|---|---|
| Definition | \( d(f(x), f(y)) \le q \, d(x, y) \) for all \( x, y \), with \( 0 \le q < 1 \); such a map is Lipschitz with modulus below 1 and is therefore uniformly continuous<sup>[1](https://www.ssc.wisc.edu/~jkennan/teaching/ContractionMappingTheorem.pdf)</sup> |
| Banach fixed point theorem | A contraction on a complete metric space has a unique fixed point, reached by iterating \( f \) from any starting point<sup>[1](https://www.ssc.wisc.edu/~jkennan/teaching/ContractionMappingTheorem.pdf)</sup> |
| Origin | Published by Stefan Banach in 1922 in *Fundamenta Mathematicae*, volume 3, pages 133–181<sup>[2](https://oamonitor.ireland.openaire.eu/national/search/publication?pid=10.4064%2Ffm-3-1-133-181)</sup> |
| A priori error bound | \( d(x_k, x_*) \le \dfrac{q^k}{1-q} \, d(x_0, x_1) \), which tells how many iterations guarantee a given accuracy<sup>[3](https://www.cs.cmu.edu/~suvrit/teach/lect22.pdf)</sup> |
| A posteriori error bound | \( d(x_{k+1}, x_*) \le \dfrac{q}{1-q} \, d(x_k, x_{k+1}) \), computable from two successive iterates<sup>[3](https://www.cs.cmu.edu/~suvrit/teach/lect22.pdf)</sup> |
| Boundary case \( q = 1 \) | Such maps are called nonexpansive; \( x \mapsto x + 1 \) is nonexpansive but has no fixed point<sup>[3](https://www.cs.cmu.edu/~suvrit/teach/lect22.pdf)</sup> |
| Main applications | Existence and uniqueness for differential equations (Picard–Lindelöf), inverse and implicit function theorems, and dynamic programming<sup>[4](https://link.springer.com/article/10.1007/s13398-024-01636-6)</sup><sup> • </sup><sup>[5](https://www2.math.upenn.edu/~gressman/analysis/12-inversefuncthm.html)</sup> |

## The theorem and its proof

The theorem is usually stated as follows. Let \( (X, d) \) be a complete metric space, meaning every [Cauchy sequence](https://www.edgepedia.org/cauchy-sequence) converges in \( X \), and let \( f : X \to X \) satisfy \( d(f(x), f(y)) \le q \, d(x, y) \) with \( q < 1 \). Then \( f \) has a unique fixed point.<sup>[1](https://www.ssc.wisc.edu/~jkennan/teaching/ContractionMappingTheorem.pdf)</sup>

**Existence by iteration.** Starting from any \( x_0 \in X \), define \( x_{n+1} = f(x_n) \). The contraction property gives \( d(x_{n+1}, x_n) \le q^n d(x_1, x_0) \), and summing these distances shows that \( d(x_{n+k}, x_n) \le \dfrac{q^n}{1-q} d(x_0, x_1) \), which tends to zero because \( 0 < q < 1 \). The sequence is therefore Cauchy, and completeness guarantees that it converges to some \( x_* \in X \); continuity of \( f \) then gives \( f(x_*) = x_* \).<sup>[1](https://www.ssc.wisc.edu/~jkennan/teaching/ContractionMappingTheorem.pdf)</sup>

**Uniqueness.** If \( x \) and \( y \) were two distinct fixed points, then \( d(x, y) = d(f(x), f(y)) \le q \, d(x, y) \), which is impossible for \( q < 1 \) unless \( d(x, y) = 0 \), that is, \( x = y \).<sup>[1](https://www.ssc.wisc.edu/~jkennan/teaching/ContractionMappingTheorem.pdf)</sup>

The proof is constructive, which distinguishes the theorem from earlier fixed point results. Brouwer's 1909–1912 theorem guarantees a fixed point for any continuous map of a closed ball in \( \mathbb{R}^n \) into itself, but it is purely existential and says nothing about how to find the point. Banach's theorem gives existence, uniqueness, and a convergent sequence of successive approximations.<sup>[4](https://link.springer.com/article/10.1007/s13398-024-01636-6)</sup>

## Error estimates and convergence rate

Convergence is linear with rate \( q \): each iteration reduces the distance to the fixed point by at least the factor \( q \), \( d(x_{k+1}, x_*) \le q \, d(x_k, x_*) \).<sup>[3](https://www.cs.cmu.edu/~suvrit/teach/lect22.pdf)</sup> Two quantitative bounds come out of the proof:<sup>[3](https://www.cs.cmu.edu/~suvrit/teach/lect22.pdf)</sup><sup> • </sup><sup>[4](https://link.springer.com/article/10.1007/s13398-024-01636-6)</sup>

- The **a priori estimate** \( d(x_k, x_*) \le \dfrac{q^k}{1-q} d(x_0, x_1) \) bounds the number of iterations needed to reach a specified accuracy before computing begins.
- The **a posteriori estimate** \( d(x_k, x_*) \le \dfrac{q}{1-q} d(x_{k-1}, x_k) \) bounds the error after each step using only the last two iterates, so if two successive iterates are nearly equal, the current iterate is close to the true fixed point.

A smaller contraction factor means faster convergence, and the number of iterations needed for a given tolerance can be read off directly from \( q \).<sup>[4](https://link.springer.com/article/10.1007/s13398-024-01636-6)</sup>

## Why the hypotheses matter

Both hypotheses, the contraction factor strictly below 1 and completeness of the space, are needed.

- **Factor \( q = 1 \) is not enough.** A map satisfying \( d(f(x), f(y)) \le d(x, y) \) is called nonexpansive. The translation \( x \mapsto x + 1 \) on the real line is nonexpansive but has no fixed point.<sup>[3](https://www.cs.cmu.edu/~suvrit/teach/lect22.pdf)</sup> A sharper counterexample is the swap map on the two-point discrete space \( X = \{0, 1\} \) with \( d(0,1) = 1 \), defined by \( T(0) = 1 \) and \( T(1) = 0 \): it satisfies the Lipschitz condition with \( c = 1 \) yet has no fixed point.<sup>[6](https://www.math.ucdavis.edu/~hunter/book/ch3.pdf)</sup>
- **Completeness is not enough without contraction.** Fixed point theorems exist for weaker maps, but they give different conclusions. The Schauder fixed point theorem guarantees a fixed point for any continuous map of a convex, compact subset of a [Banach space](https://www.edgepedia.org/banach-space) into itself, but it does not give uniqueness, and its proof is topological rather than iterative.<sup>[6](https://www.math.ucdavis.edu/~hunter/book/ch3.pdf)</sup>
- **Completeness is not strictly necessary.** There are incomplete metric spaces on which every contraction nevertheless has a fixed point, so completeness is a sufficient rather than a necessary condition.<sup>[7](https://arxiv.org/abs/1508.05173)</sup>

Some authors use "contraction" for the case \( c = 1 \) and "strict contraction" for \( c < 1 \); the fixed point theorem requires the strict case.<sup>[6](https://www.math.ucdavis.edu/~hunter/book/ch3.pdf)</sup>

## History

[Stefan Banach](https://www.edgepedia.org/stefan-banach) (1892–1945) published the theorem in 1922 in *Fundamenta Mathematicae*, in the paper *Sur les opérations dans les ensembles abstraits et leur application aux équations intégrales*, which formed part of his doctoral work at the Jan Kazimierz University in Lwów.<sup>[2](https://oamonitor.ireland.openaire.eu/national/search/publication?pid=10.4064%2Ffm-3-1-133-181)</sup><sup> • </sup><sup>[4](https://link.springer.com/article/10.1007/s13398-024-01636-6)</sup> The 1922 paper also introduced the axioms of what are now called Banach spaces, and the fixed point theorem was proved in that setting.<sup>[4](https://link.springer.com/article/10.1007/s13398-024-01636-6)</sup> The result is sometimes called the Banach–[Picard theorem](https://www.edgepedia.org/picard-theorem), reflecting its connection with Picard's earlier iteration method.<sup>[8](https://studyres.com/doc/3119505/fixed-point-theorems--supplementary-notes-appm)</sup>

## Extensions

The contraction condition is restrictive, and much of metric fixed point theory consists of weakenings of it. The best known is Kannan's theorem of 1968: if a self-map \( T \) of a complete metric space satisfies

\[ d(T(x), T(y)) \le r \big[ d(T(x), x) + d(T(y), y) \big] \]

for some \( r \in (0, \tfrac{1}{2}) \), then \( T \) has a unique fixed point. A Kannan contraction need not be continuous, so the class is not contained in the Banach class. The converse also holds: if every Kannan contraction with a fixed \( r \in (0, \tfrac{1}{2}) \) has a fixed point, the space is complete, a result due to Papagudi V. Subrahmanyam. At \( r = \tfrac{1}{2} \) the conclusion fails, and a Kannan contraction may have no fixed point even if it is continuous.<sup>[4](https://link.springer.com/article/10.1007/s13398-024-01636-6)</sup> Further variants include monotone contractions in uniformly convex smooth Banach spaces and conditions comparing Banach- and Kannan-type mappings along Picard sequences.<sup>[9](https://onlinelibrary.wiley.com/doi/10.1155/2020/2879283)</sup><sup> • </sup><sup>[10](https://arxiv.org/abs/2605.11883)</sup>

## Applications

**Differential equations.** The Picard–Lindelöf theorem, which gives existence and uniqueness of a local solution to an initial value problem \( \dot x = f(t, x) \) when \( f \) is locally Lipschitz in space, is proved by rewriting the ODE as an integral equation \( x(t) = x_0 + \int_{t_0}^t f(s, x(s)) \, ds \) and showing that the integral operator is a contraction on a complete space of continuous functions for a small enough time interval.<sup>[11](https://www.math.cmu.edu/~gautam/c/2026-632/notes/existence.html)</sup> The Picard–Lindelöf theorem actually predates Banach's paper by 32 years, and its original proof did not use the contraction principle; it is now usually presented as an application of it.<sup>[4](https://link.springer.com/article/10.1007/s13398-024-01636-6)</sup>

**Inverse and implicit function theorems.** The equation \( f(x) = a \) can be reformulated as a fixed point equation and solved with the contraction mapping principle, which is the route used to prove the inverse function theorem in courses on advanced analysis.<sup>[5](https://www2.math.upenn.edu/~gressman/analysis/12-inversefuncthm.html)</sup>

**Dynamic programming and reinforcement learning.** In dynamic programming, the Bellman operator maps value functions to value functions, and its fixed point is the optimal value function. When the operator is a contraction with factor below 1 on a complete space, Banach's theorem guarantees that value iteration converges to the optimal values regardless of the starting function.<sup>[12](https://upcommons.upc.edu/bitstream/handle/2117/95729/TGE1de1.pdf)</sup><sup> • </sup><sup>[13](https://people.anu.edu.au/timothy.kam/work/teaching/econ4422/html/dynamic-programming.html)</sup>

**Inverse problems.** Banach's theorem for contraction maps is used to solve inverse problems for ordinary differential equations, for example finding a functional form \( \dot x(t) = f(x, t) \), restricted to a class such as affine or quadratic forms, that admits a given data function as an exact or approximate solution.<sup>[14](https://doi.org/10.1088/0266-5611/15/3/308)</sup>

## References

1. [Contraction Mapping Theorem (Banach Fixed Point Theorem), University of Wisconsin lecture notes](https://www.ssc.wisc.edu/~jkennan/teaching/ContractionMappingTheorem.pdf)
2. [Stefan Banach (1922). Sur les opérations dans les ensembles abstraits et leur application aux équations intégrales. Fundamenta Mathematicae 3, 133–181.](https://oamonitor.ireland.openaire.eu/national/search/publication?pid=10.4064%2Ffm-3-1-133-181)
3. [Advanced Optimization lecture notes, Banach contraction, Carnegie Mellon University](https://www.cs.cmu.edu/~suvrit/teach/lect22.pdf)
4. [The Banach Fixed Point Theorem: selected topics from its hundred-year history. RACSAM (2024).](https://link.springer.com/article/10.1007/s13398-024-01636-6)
5. [The Inverse Function Theorem, Advanced Analysis, University of Pennsylvania](https://www2.math.upenn.edu/~gressman/analysis/12-inversefuncthm.html)
6. [The Contraction Mapping Theorem, chapter 3 of Hunter's Applied Analysis, UC Davis](https://www.math.ucdavis.edu/~hunter/book/ch3.pdf)
7. [Fixed points and completeness in metric and in generalized metric spaces, arXiv](https://arxiv.org/abs/1508.05173)
8. [Fixed Point Theorems, supplementary notes APPM](https://studyres.com/doc/3119505/fixed-point-theorems--supplementary-notes-appm)
9. [The Monotone Contraction Mapping Theorem, Journal of Mathematics (2020)](https://onlinelibrary.wiley.com/doi/10.1155/2020/2879283)
10. [A comparison of the weakest contractive conditions for Banach and Kannan mappings, arXiv](https://arxiv.org/abs/2605.11883)
11. [Existence and Uniqueness, lecture notes, Carnegie Mellon University](https://www.math.cmu.edu/~gautam/c/2026-632/notes/existence.html)
12. [Dynamic Programming, UPC thesis chapter](https://upcommons.upc.edu/bitstream/handle/2117/95729/TGE1de1.pdf)
13. [Dynamic Programming, Dynamic Macroeconomics course notes, ANU](https://people.anu.edu.au/timothy.kam/work/teaching/econ4422/html/dynamic-programming.html)
14. [Solving inverse problems for ordinary differential equations using the Picard contraction mapping, Inverse Problems (1998)](https://doi.org/10.1088/0266-5611/15/3/308)

---
*Topic: Encyclopedia › Physical world and mathematics › Mathematics and statistics › Analysis and mathematical models › Real analysis*

*Initially written Sep 17, 2026 · Reviewed: Oct 15, 2026 · Edited: Oct 15, 2026 · Last review: Oct 15, 2026*

*Copyright 2026 EdgeChat AI, a subsidiary of Biostate AI.*

License: Edgepedia Community License 1.0, https://www.edgepedia.org/edgepedia/license