Ritt Operators in Banach Spaces

• Ritt operators are bounded operators on Banach spaces with power-bounded iterates and discrete derivative decay at an order of 1/n, characterized by specific spectral and resolvent conditions.
• They support a bounded H∞-functional calculus on Stolz domains, establishing an exact parallel with sectorial operators and analytic semigroups in continuous theory.
• Ritt operators find practical applications in discrete evolution equations, numerical schemes, and ergodic theory, where resolvent control guarantees stability and convergence.

A Ritt operator on a complex Banach space $X$ is a bounded operator $T$ whose iterates are power-bounded and whose discrete derivative decays at order $1/n$: the sets $\{T^n:n\ge 0\}$ and $\{n(T^n-T^{n-1}):n\ge 1\}$ are bounded. Equivalently, $\sigma(T)\subset \overline{\mathbb D}$ and $\sup_{|z|>1}\|(z-1)R(z,T)\|<\infty$, or, with $A=I-T$, the operator $A$ is sectorial of type $<\pi/2$. In this sense, Ritt operators are the discrete-time analogues of sectorial operators and generators of bounded analytic semigroups: powers $T$0 replace $T$1, Stolz domains replace sectors, and discrete differences replace time derivatives (Merdy, 2012).

1. Definition, resolvent geometry, and the discrete-time viewpoint

The standard definition is resolvent-free: $T$2 is Ritt when both $T$3 and $T$4 are bounded. This is equivalent to the spectral-resolvent condition

$T$5

and also to the statement that $T$6 is sectorial of type $T$7 (Merdy, 2012). In the terminology of Tadmor–Ritt theory, the quantity

$T$8

is the Ritt constant (Schwenninger, 2015).

The natural spectral regions for Ritt operators are Stolz domains. For $T$9, the Stolz domain $1/n$0 is the interior of the convex hull of the point $1/n$1 and the disc $1/n$2. If $1/n$3 is Ritt, then there exists $1/n$4 such that $1/n$5, and for any $1/n$6, the family $1/n$7 is bounded (Merdy, 2012). The basic geometric estimate

$1/n$8

encodes the tangential approach to the boundary point $1/n$9 (Arhancet et al., 2015).

The continuous/discrete analogy is structural rather than heuristic.

Continuous theory	Discrete theory
Sectorial operator $\{T^n:n\ge 0\}$0	Ritt operator $\{T^n:n\ge 0\}$1
Analytic semigroup $\{T^n:n\ge 0\}$2	Powers $\{T^n:n\ge 0\}$3
Sectors $\{T^n:n\ge 0\}$4	Stolz domains $\{T^n:n\ge 0\}$5
$\{T^n:n\ge 0\}$6	$\{T^n:n\ge 0\}$7

This analogy is exact enough that many sectorial arguments transfer to Ritt operators after replacing $\{T^n:n\ge 0\}$8 by $\{T^n:n\ge 0\}$9 and sectors by Stolz domains (Merdy, 2012). A common misconception is that power-boundedness alone is close to the Ritt property; it is not. Rittness requires the additional discrete derivative bound, and the latter is the genuinely analytic component.

2. $\{n(T^n-T^{n-1}):n\ge 1\}$0-functional calculus on Stolz domains

For a Ritt operator $\{n(T^n-T^{n-1}):n\ge 1\}$1 of type $\{n(T^n-T^{n-1}):n\ge 1\}$2 and $\{n(T^n-T^{n-1}):n\ge 1\}$3, the holomorphic functional calculus is defined on the algebra $\{n(T^n-T^{n-1}):n\ge 1\}$4 of bounded holomorphic functions $\{n(T^n-T^{n-1}):n\ge 1\}$5 on $\{n(T^n-T^{n-1}):n\ge 1\}$6 satisfying a vanishing condition at $\{n(T^n-T^{n-1}):n\ge 1\}$7, typically

$\{n(T^n-T^{n-1}):n\ge 1\}$8

for some $\{n(T^n-T^{n-1}):n\ge 1\}$9. One sets

$\sigma(T)\subset \overline{\mathbb D}$0

and says that $\sigma(T)\subset \overline{\mathbb D}$1 has a bounded $\sigma(T)\subset \overline{\mathbb D}$2-functional calculus if

$\sigma(T)\subset \overline{\mathbb D}$3

for all $\sigma(T)\subset \overline{\mathbb D}$4 (Merdy, 2012). Testing on polynomials is sufficient: bounded $\sigma(T)\subset \overline{\mathbb D}$5-calculus is equivalent to the same estimate for all polynomials (Merdy, 2012).

The decisive transfer principle is that bounded Stolz-domain calculus for $\sigma(T)\subset \overline{\mathbb D}$6 is equivalent to bounded sectorial calculus for $\sigma(T)\subset \overline{\mathbb D}$7: $\sigma(T)\subset \overline{\mathbb D}$8 admits a bounded $\sigma(T)\subset \overline{\mathbb D}$9-calculus for some $\sup_{|z|>1}\|(z-1)R(z,T)\|<\infty$0 if and only if $\sup_{|z|>1}\|(z-1)R(z,T)\|<\infty$1 admits a bounded $\sup_{|z|>1}\|(z-1)R(z,T)\|<\infty$2-calculus for some $\sup_{|z|>1}\|(z-1)R(z,T)\|<\infty$3 (Merdy, 2012). This is the main bridge between discrete and continuous theories.

Quantitative estimates are also available. For a Tadmor–Ritt operator $\sup_{|z|>1}\|(z-1)R(z,T)\|<\infty$4 and a polynomial $\sup_{|z|>1}\|(z-1)R(z,T)\|<\infty$5, Schwenninger proved

$\sup_{|z|>1}\|(z-1)R(z,T)\|<\infty$6

with absolute constants $\sup_{|z|>1}\|(z-1)R(z,T)\|<\infty$7. In particular,

$\sup_{|z|>1}\|(z-1)R(z,T)\|<\infty$8

which recovers the best known power-bound asymptotics $\sup_{|z|>1}\|(z-1)R(z,T)\|<\infty$9 from the functional calculus itself (Schwenninger, 2015).

These estimates clarify the role of the resolvent constant: $A=I-T$0 governs not only stability of powers but also the size of the discrete holomorphic calculus. A plausible implication is that, in applications, resolvent control near $A=I-T$1 is the correct proxy for discrete analytic regularity.

3. Square functions and discrete Littlewood–Paley theory

For $A=I-T$2, $A=I-T$3, a basic square function is

$A=I-T$4

More generally, for integer $A=I-T$5,

$A=I-T$6

and on $A=I-T$7-spaces this is equivalent to the corresponding $A=I-T$8-square function by Khintchine–Kahane inequalities (Merdy, 2012).

Le Merdy and Xu established the discrete analogue of the Cowling–Doust–McIntosh–Yagi theorem: for a Ritt operator $A=I-T$9, bounded $A$0-calculus is equivalent to the simultaneous square function estimates for $A$1 and $A$2. Precisely, $A$3 admits a bounded $A$4-calculus for some $A$5 if and only if

$A$6

for all $A$7, $A$8 (Merdy, 2012). On Hilbert space, this is further equivalent to similarity to a contraction, under the Ritt assumption (Merdy, 2012).

For $A$9-Ritt operators, the square-function scale is stable under change of exponent. If $<\pi/2$0 is $<\pi/2$1-Ritt on $<\pi/2$2, then for any $<\pi/2$3,

$<\pi/2$4

and the same phenomenon holds on reflexive Banach spaces with finite cotype (Arhancet et al., 2011). This equivalence is technically important because it allows one to choose the exponent best adapted to the argument.

The two-sided nature of the criterion is essential. On Hilbert space there exist Ritt operators $<\pi/2$5 with $<\pi/2$6 that are not similar to contractions because the adjoint estimate fails (Merdy, 2012). This rules out the common shortcut that one-sided square-function control should suffice.

Recent endpoint work extends the theory to $<\pi/2$7. If $<\pi/2$8 is a Ritt operator on $<\pi/2$9, then the generalized square function

$T$00

is bounded on $T$01 whenever $T$02 (Hults et al., 2023). The 2025 continuation adds weak type $T$03 results in a convolution setting and further variational and oscillation estimates at the endpoint (Hults et al., 9 Jul 2025).

4. Dilations, contractive models, and similarity

A central structural theme is that good functional calculus is equivalent to dilation into a better behaved model operator. For Ritt operators on reflexive Banach spaces $T$04 such that $T$05 and $T$06 have finite cotype, bounded $T$07-calculus yields an isometric dilation: there exist a measure space $T$08, an isometric isomorphism

$T$09

and bounded operators

$T$10

such that

$T$11

If $T$12 is ordered, $T$13 can be chosen positive (Arhancet et al., 2015).

On UMD spaces, the picture sharpens: a Ritt operator $T$14 has bounded $T$15-calculus if and only if it is the compression of a contractive Ritt operator $T$16 on some $T$17, with

$T$18

On scalar $T$19, $T$20 may be chosen contractive and positive (Arhancet et al., 2015). Earlier work formulated an equivalent “loose dilation” criterion on $T$21: $T$22 and $T$23 satisfy square function estimates if and only if $T$24 is $T$25-Ritt and there exist $T$26 with bounded $T$27 such that $T$28 for all $T$29 (Arhancet et al., 2011).

These results are discrete analogues of dilation theorems for analytic semigroups. They supply functional models in which $T$30 is studied through shifts or isometries on larger $T$31-spaces. In uniformly convex spaces, bounded $T$32-calculus even yields an equivalent uniformly convex norm for which $T$33 becomes a contraction (Arhancet et al., 2015).

On Hilbert space, the similarity problem is especially clean: a power-bounded operator $T$34 is a Ritt operator with bounded $T$35-calculus if and only if $T$36 is Ritt and similar to a contraction (Merdy, 2012). The discrete dilation theory thus recovers a classical operator-theoretic paradigm—contraction models—but only after the Ritt condition isolates the analytic regime.

5. Multivariable, noncommutative, and generalized Ritt theories

The one-variable theory extends to commuting families. For a commuting $T$37-tuple $T$38 of Ritt operators, a joint $T$39-functional calculus is defined on products $T$40 by iterated Dunford–Riesz integrals (Arrigoni et al., 2019). Under geometric assumptions such as “$T$41 is a Banach lattice” or “$T$42 or $T$43 has property $T$44”, a commuting tuple has a joint calculus if and only if each coordinate does (Arrigoni et al., 2019). On $T$45, joint calculus is equivalent to joint dilation to commuting positive contractive Ritt operators, and on UMD spaces with property $T$46 it is equivalent to joint dilation to commuting contractive Ritt operators on vector-valued $T$47-spaces (Arrigoni et al., 2019). Parallel equivalences between joint calculus, joint loose dilations, and joint polynomial boundedness were developed for commuting $T$48-tuples on $T$49-spaces (Mohanty et al., 2017).

Square-function methods also admit a multivariable form. For a commuting $T$50-tuple of $T$51-Ritt operators, joint $T$52-calculus implies multivariable square function estimates, and conversely, under $T$53-Ritt and suitable geometry such as $T$54-convexity or property $T$55, square function estimates yield dilations into $T$56-tuples of isomorphisms on Bochner spaces with $T$57-bounded calculus (Arrigoni, 2020). On Hilbert space, joint $T$58-calculus for commuting Ritt tuples is equivalent to joint similarity to a commuting tuple of contractions (Arrigoni et al., 2019).

In noncommutative $T$59-spaces, the theory splits into column and row structures. Completely bounded $T$60-calculus implies both Col-Ritt and Row-Ritt properties, and square functions must be formulated in column and row forms rather than through a single scalar $T$61-norm (Arhancet, 2011). A notable phenomenon is that column and row square functions need not be equivalent even when the operator admits a completely bounded $T$62-calculus; explicit counterexamples are given on Schatten classes (Arhancet, 2011). For $T$63, however, there is a decomposition result: every $T$64 can be written as $T$65 with controlled column square function of $T$66 and row square function of $T$67 (Arhancet, 2011).

Several generalizations enlarge the class itself. $T$68-Ritt operators are discrete analogues of $T$69-sectorial operators and carry an $T$70-calculus on $T$71-Stolz domains, together with a transference principle between $T$72 and $T$73 (Ray, 2017). Ritt$T$74 operators replace the distinguished boundary point $T$75 by a finite set $T$76, with generalized Stolz domains $T$77, adapted square functions, and equivalences between bounded calculus, quadratic calculus, and square function estimates under finite cotype hypotheses (Bouabdillah, 2024).

6. Applications, subordination, quantitative bounds, and counterexamples

Ritt operators arise naturally in discrete evolution equations, numerical schemes, and ergodic theory. They model time-stepping operators for parabolic problems, and bounded $T$78-calculus is closely related to discrete maximal regularity (Arhancet et al., 2015). In ergodic theory they control convergence rates of iterates and ergodic averages, especially when positivity or contractivity is available (Arhancet et al., 2015).

Discrete subordination is a major permanence mechanism. If

$T$79

and $T$80 is Ritt, then $T$81 is again Ritt; more precisely, convex combinations of powers of a Ritt operator are Ritt and preserve Stolz type (Gomilko et al., 2015). For regular Hausdorff functions, the improving property is characterized geometrically: $T$82 is Ritt for every power-bounded $T$83 if and only if $T$84 is contained in a sector $T$85 (Gomilko et al., 2015). This extends Dungey’s discrete subordination program.

Subordinated operators from group representations provide another source. For a bounded strongly continuous representation $T$86 of a locally compact abelian group and a probability measure $T$87, the averaged operator

$T$88

is Ritt with bounded $T$89-functional calculus whenever $T$90 is a UMD Banach lattice and $T$91 has bounded angular ratio. If $T$92 is the square of a symmetric probability measure and $T$93 is $T$94-convex, then $T$95 is Ritt. The latter statement fails on any non-$T$96-convex Banach space (Lancien et al., 2017).

The theory also has sharp negative results. Not every Ritt operator is $T$97-Ritt, and on a Banach space with a Schauder decomposition that is not $T$98-Schauder, there exists a Ritt multiplier $T$99 such that the family $1/n$00 is not $1/n$01-bounded (Arnold et al., 2018). Thus the passage from boundedness to $1/n$02-boundedness is genuinely geometric and cannot be taken for granted. On Hilbert space, one-sided square-function bounds do not force similarity to a contraction because the adjoint estimate may fail (Merdy, 2012). In the multivariable setting, the general von Neumann inequality for commuting contractions remains open in dimension $1/n$03, although Ritt hypotheses permit partial positive results through joint similarity and dilation theorems (Arrigoni et al., 2019).

Taken together, these developments establish Ritt operators as the canonical discrete counterpart of sectorial operators: they support a holomorphic functional calculus on Stolz domains, admit square-function and dilation characterizations, persist under discrete subordination, and extend to multivariable, noncommutative, and generalized boundary-point settings. The subject now connects discrete maximal regularity, ergodic theory, harmonic analysis, Banach space geometry, and operator dilation theory in a single analytic framework.