Gevrey-Modulation Spaces

• Gevrey-modulation spaces are modulation spaces defined with Gevrey-type exponential weights (e^(|ξ|^(1/s)) where s>1), offering a refined framework beyond classical Sobolev scales.
• They exhibit a Banach algebra structure with stability under pointwise multiplication and support analytic and non-analytic superposition operators, making them useful in nonlinear PDE analysis.
• The spaces incorporate Gelfand–Shilov window functions and frequency-uniform decomposition, enabling precise microlocal control and embedding into spaces of smooth functions.

Gevrey-modulation spaces are modulation spaces with Gevrey-type exponential weights on the frequency side, typically of the form $e^{|\xi|^{1/s}}$ with $s>1$. They were introduced in the frequency-uniform decomposition framework as a way of “leaving the Sobolev frame and proceeding to the Gevrey frame,” so that the stronger high-frequency control could support nonlinear operations that are delicate or unavailable in polynomially weighted settings (Reich, 2014). In later work, the same scale was developed systematically under the notation $\mathcal{GM}_{p,q}^s$, together with analytic and non-analytic superposition results and an extension to a weaker ultra-differentiable modulation scale (Reich et al., 2015).

1. Definitions and basic framework

The starting point is the short-time Fourier transform

$$V_\phi f(x,\xi)=(2\pi)^{-n/2}\int_{\mathbb R^n} f(s)\,\overline{\phi(s-x)}\,e^{-is\cdot \xi}\,ds,$$

with a nonzero window $\phi$. Classical weighted modulation spaces use polynomial weights such as

$\langle \xi\rangle^s=(1+|\xi|^2)^{s/2},$

and may be written either through the STFT or through a frequency-uniform decomposition. If $(\sigma_k)_{k\in\mathbb Z^n}$ is a partition of unity on frequency space, the associated localization operators are

$$\square_k:=\mathcal F^{-1}\big(\sigma_k\mathcal F(\cdot)\big).$$

Then

$$M_s^{p,q}(\mathbb R^n)=\left\{f\in\mathcal S'(\mathbb R^n):\left(\sum_{k\in\mathbb Z^n}\langle k\rangle^{sq}\|\square_k f\|_{L^p}^q\right)^{1/q}<\infty\right\}.$$

This is the “Sobolev frame” formulation emphasized in (Reich, 2014).

The Gevrey-modulation space replaces the polynomial frequency weight by the subexponential weight $e^{|\xi|^{1/s}}$. In the notation of (Reich, 2014), for $s>1$0, $s>1$1, and a fixed nonzero window $s>1$2,

$s>1$3

with norm

$s>1$4

The equivalent discrete norm is

$s>1$5

In (Reich et al., 2015) the same class is written $s>1$6, with the same weight $s>1$7. That paper defines the space for $s>1$8, but the main algebra and superposition results require $s>1$9 (Reich et al., 2015).

The natural window class is no longer the Schwartz space alone, but the Gelfand–Shilov class. For $\mathcal{GM}_{p,q}^s$0 and $\mathcal{GM}_{p,q}^s$1,

$\mathcal{GM}_{p,q}^s$2

and

$\mathcal{GM}_{p,q}^s$3

This is the window class used in (Reich, 2014) precisely because Gevrey weights are too strong for the Schwartz class to be the natural ambient class.

2. Structural properties and relation to Gevrey regularity

The basic structure of Gevrey-modulation spaces follows the standard modulation-space pattern: they are Banach spaces, and the STFT definition is equivalent to the frequency-uniform decomposition definition. In (Reich, 2014) this is stated by transferring the usual arguments for classical modulation spaces to the Gevrey-weighted setting, so the resulting space is independent of the particular admissible window or partition of unity (Reich, 2014).

A central background fact is the Gelfand–Shilov characterization

$\mathcal{GM}_{p,q}^s$4

together with the STFT characterization

$\mathcal{GM}_{p,q}^s$5

These facts explain why $\mathcal{GM}_{p,q}^s$6-windows are built into the definition of $\mathcal{GM}_{p,q}^s$7 (Reich, 2014).

The frequency weight $\mathcal{GM}_{p,q}^s$8 is much stronger than any polynomial weight: $\mathcal{GM}_{p,q}^s$9 Accordingly, Gevrey-modulation spaces encode regularity beyond Sobolev scales. A useful $$V_\phi f(x,\xi)=(2\pi)^{-n/2}\int_{\mathbb R^n} f(s)\,\overline{\phi(s-x)}\,e^{-is\cdot \xi}\,ds,$$0 model is provided in (Reich et al., 2015): $$V_\phi f(x,\xi)=(2\pi)^{-n/2}\int_{\mathbb R^n} f(s)\,\overline{\phi(s-x)}\,e^{-is\cdot \xi}\,ds,$$1 This gives a direct Fourier-side interpretation of the space as a subexponentially weighted $$V_\phi f(x,\xi)=(2\pi)^{-n/2}\int_{\mathbb R^n} f(s)\,\overline{\phi(s-x)}\,e^{-is\cdot \xi}\,ds,$$2 class (Reich et al., 2015).

The spaces also embed into bounded functions. In (Reich, 2014), Corollary 3.9 states

$$V_\phi f(x,\xi)=(2\pi)^{-n/2}\int_{\mathbb R^n} f(s)\,\overline{\phi(s-x)}\,e^{-is\cdot \xi}\,ds,$$3

A related regularity consequence is stronger: if $$V_\phi f(x,\xi)=(2\pi)^{-n/2}\int_{\mathbb R^n} f(s)\,\overline{\phi(s-x)}\,e^{-is\cdot \xi}\,ds,$$4, then $$V_\phi f(x,\xi)=(2\pi)^{-n/2}\int_{\mathbb R^n} f(s)\,\overline{\phi(s-x)}\,e^{-is\cdot \xi}\,ds,$$5, and there exists $$V_\phi f(x,\xi)=(2\pi)^{-n/2}\int_{\mathbb R^n} f(s)\,\overline{\phi(s-x)}\,e^{-is\cdot \xi}\,ds,$$6 such that

$$V_\phi f(x,\xi)=(2\pi)^{-n/2}\int_{\mathbb R^n} f(s)\,\overline{\phi(s-x)}\,e^{-is\cdot \xi}\,ds,$$7

Hence

$$V_\phi f(x,\xi)=(2\pi)^{-n/2}\int_{\mathbb R^n} f(s)\,\overline{\phi(s-x)}\,e^{-is\cdot \xi}\,ds,$$8

in the classical Gevrey sense (Reich et al., 2015).

A common misconception is that these spaces coincide with the usual Gevrey classes. The literature states the opposite more carefully: Gevrey-modulation spaces are not identical to standard Gevrey classes, but they are modulation-space analogues with Gevrey-type Fourier weights and $$V_\phi f(x,\xi)=(2\pi)^{-n/2}\int_{\mathbb R^n} f(s)\,\overline{\phi(s-x)}\,e^{-is\cdot \xi}\,ds,$$9-$\phi$0 microlocal control (Reich, 2014).

3. Multiplication and Banach algebra structure

The first major nonlinear property is closure under pointwise multiplication. In (Reich, 2014), Theorem 3.12 states that if $\phi$1, $\phi$2, and $\phi$3, then

$\phi$4

and

$\phi$5

In (Reich et al., 2015) this is sharpened to the more flexible statement: if $\phi$6, $\phi$7, and

$\phi$8

then

$\phi$9

with

$\langle \xi\rangle^s=(1+|\xi|^2)^{s/2},$0

As a consequence, for every $\langle \xi\rangle^s=(1+|\xi|^2)^{s/2},$1 and $\langle \xi\rangle^s=(1+|\xi|^2)^{s/2},$2,

$\langle \xi\rangle^s=(1+|\xi|^2)^{s/2},$3

is a Banach algebra under pointwise multiplication (Reich et al., 2015).

The decisive estimate is the Gevrey weight inequality. In the formulation of (Reich et al., 2015), if $\langle \xi\rangle^s=(1+|\xi|^2)^{s/2},$4 and

$\langle \xi\rangle^s=(1+|\xi|^2)^{s/2},$5

then

$\langle \xi\rangle^s=(1+|\xi|^2)^{s/2},$6

In (Reich, 2014) this appears in closely related form as

$\langle \xi\rangle^s=(1+|\xi|^2)^{s/2},$7

The extra decaying factor

$\langle \xi\rangle^s=(1+|\xi|^2)^{s/2},$8

is what yields summability on the lattice and makes the nonlinear argument work (Reich, 2014).

The proof uses the frequency-uniform decomposition

$\langle \xi\rangle^s=(1+|\xi|^2)^{s/2},$9

together with support properties of $(\sigma_k)_{k\in\mathbb Z^n}$0, Hölder’s inequality, Nikol’skij’s inequality, and summability of the exponentially decaying remainder (Reich, 2014). The restriction $(\sigma_k)_{k\in\mathbb Z^n}$1 is essential rather than cosmetic: (Reich et al., 2015) explicitly notes that $(\sigma_k)_{k\in\mathbb Z^n}$2 is not an algebra.

4. Superposition and non-analytic composition

The second major nonlinear property is superposition. The algebra estimate already implies the analytic functional calculus: if

$(\sigma_k)_{k\in\mathbb Z^n}$3

is analytic near $(\sigma_k)_{k\in\mathbb Z^n}$4 with $(\sigma_k)_{k\in\mathbb Z^n}$5, then $(\sigma_k)_{k\in\mathbb Z^n}$6 implies $(\sigma_k)_{k\in\mathbb Z^n}$7, because

$(\sigma_k)_{k\in\mathbb Z^n}$8

(Reich et al., 2015).

The distinctive result is the non-analytic superposition theorem. In (Reich, 2014), Theorem 3.19 assumes that $(\sigma_k)_{k\in\mathbb Z^n}$9 and that $$\square_k:=\mathcal F^{-1}\big(\sigma_k\mathcal F(\cdot)\big).$$0 is a complex measure on $$\square_k:=\mathcal F^{-1}\big(\sigma_k\mathcal F(\cdot)\big).$$1 satisfying

$$\square_k:=\mathcal F^{-1}\big(\sigma_k\mathcal F(\cdot)\big).$$2

together with

$$\square_k:=\mathcal F^{-1}\big(\sigma_k\mathcal F(\cdot)\big).$$3

If

$$\square_k:=\mathcal F^{-1}\big(\sigma_k\mathcal F(\cdot)\big).$$4

then $$\square_k:=\mathcal F^{-1}\big(\sigma_k\mathcal F(\cdot)\big).$$5, and the composition operator

$$\square_k:=\mathcal F^{-1}\big(\sigma_k\mathcal F(\cdot)\big).$$6

maps $$\square_k:=\mathcal F^{-1}\big(\sigma_k\mathcal F(\cdot)\big).$$7 into itself (Reich, 2014).

In (Reich et al., 2015), the corresponding theorem is stated for $$\square_k:=\mathcal F^{-1}\big(\sigma_k\mathcal F(\cdot)\big).$$8, $$\square_k:=\mathcal F^{-1}\big(\sigma_k\mathcal F(\cdot)\big).$$9, and $$M_s^{p,q}(\mathbb R^n)=\left\{f\in\mathcal S'(\mathbb R^n):\left(\sum_{k\in\mathbb Z^n}\langle k\rangle^{sq}\|\square_k f\|_{L^p}^q\right)^{1/q}<\infty\right\}.$$0: if the same measure condition holds and $$M_s^{p,q}(\mathbb R^n)=\left\{f\in\mathcal S'(\mathbb R^n):\left(\sum_{k\in\mathbb Z^n}\langle k\rangle^{sq}\|\square_k f\|_{L^p}^q\right)^{1/q}<\infty\right\}.$$1, then

$$M_s^{p,q}(\mathbb R^n)=\left\{f\in\mathcal S'(\mathbb R^n):\left(\sum_{k\in\mathbb Z^n}\langle k\rangle^{sq}\|\square_k f\|_{L^p}^q\right)^{1/q}<\infty\right\}.$$2

(Reich et al., 2015). The condition $$M_s^{p,q}(\mathbb R^n)=\left\{f\in\mathcal S'(\mathbb R^n):\left(\sum_{k\in\mathbb Z^n}\langle k\rangle^{sq}\|\square_k f\|_{L^p}^q\right)^{1/q}<\infty\right\}.$$3 is used to write

$$M_s^{p,q}(\mathbb R^n)=\left\{f\in\mathcal S'(\mathbb R^n):\left(\sum_{k\in\mathbb Z^n}\langle k\rangle^{sq}\|\square_k f\|_{L^p}^q\right)^{1/q}<\infty\right\}.$$4

so that the nonlinear estimate is centered at $$M_s^{p,q}(\mathbb R^n)=\left\{f\in\mathcal S'(\mathbb R^n):\left(\sum_{k\in\mathbb Z^n}\langle k\rangle^{sq}\|\square_k f\|_{L^p}^q\right)^{1/q}<\infty\right\}.$$5.

The core estimate concerns $$M_s^{p,q}(\mathbb R^n)=\left\{f\in\mathcal S'(\mathbb R^n):\left(\sum_{k\in\mathbb Z^n}\langle k\rangle^{sq}\|\square_k f\|_{L^p}^q\right)^{1/q}<\infty\right\}.$$6. In (Reich, 2014), Lemma 3.17 gives

$$M_s^{p,q}(\mathbb R^n)=\left\{f\in\mathcal S'(\mathbb R^n):\left(\sum_{k\in\mathbb Z^n}\langle k\rangle^{sq}\|\square_k f\|_{L^p}^q\right)^{1/q}<\infty\right\}.$$7

with constants $$M_s^{p,q}(\mathbb R^n)=\left\{f\in\mathcal S'(\mathbb R^n):\left(\sum_{k\in\mathbb Z^n}\langle k\rangle^{sq}\|\square_k f\|_{L^p}^q\right)^{1/q}<\infty\right\}.$$8 independent of $$M_s^{p,q}(\mathbb R^n)=\left\{f\in\mathcal S'(\mathbb R^n):\left(\sum_{k\in\mathbb Z^n}\langle k\rangle^{sq}\|\square_k f\|_{L^p}^q\right)^{1/q}<\infty\right\}.$$9. The same estimate appears in (Reich et al., 2015) for $e^{|\xi|^{1/s}}$0. This bound is combined with Bochner integration in the Banach space $e^{|\xi|^{1/s}}$1 or $e^{|\xi|^{1/s}}$2 to obtain the superposition theorem (Reich, 2014).

The result is termed non-analytic because it does not assume a convergent power series for the outer function. Instead, the outer function is recovered as the inverse Fourier transform of a measure with the explicit integrability condition (31). The papers present this as a genuine non-analytic Nemytskii theorem (Reich, 2014).

5. Operator theory, periodic theory, and related modulation-space developments

Later work placed Gevrey-modulation spaces inside a broader operator-theoretic framework. One direction is continuity of pseudo-differential operators with Gevrey-Hörmander symbols on weighted modulation spaces. In (Toft, 2017), if $e^{|\xi|^{1/s}}$3, $e^{|\xi|^{1/s}}$4, $e^{|\xi|^{1/s}}$5, and $e^{|\xi|^{1/s}}$6 is an invariant BF-space, then

$e^{|\xi|^{1/s}}$7

is continuous. The same paper identifies Gevrey symbol classes as unions and intersections of modulation spaces with subexponential frequency weights: $e^{|\xi|^{1/s}}$8 where

$e^{|\xi|^{1/s}}$9

This gives a direct time-frequency realization of Gevrey-Hörmander regularity (Toft, 2017).

An anisotropic version appears in (Abdeljawad et al., 2018). There the symbol classes

$s>1$00

are controlled by

$s>1$01

and the weights satisfy

$s>1$02

Under these assumptions,

$s>1$03

is continuous (Abdeljawad et al., 2018).

A second direction is lifting and isomorphism theory. In (Abdeljawad et al., 2017), pseudo-differential operators with Gevrey-type symbols are used to construct explicit inverses

$s>1$04

with $s>1$05 in a Gevrey-weighted symbol class and $s>1$06 in the reciprocal one, and to deduce isomorphisms between weighted modulation spaces. The most concrete statement is that, for weight functions $s>1$07 moderated by GRS submultiplicative weights,

$s>1$08

is an isomorphism from

$s>1$09

for every $s>1$10 (Abdeljawad et al., 2017).

Bilinear pseudo-differential operators with Gevrey-Hörmander symbols were treated in (Abdeljawad et al., 2019). That paper does not introduce a space literally named “Gevrey-modulation space,” but it gives STFT characterizations of the relevant symbol classes through estimates such as

$s>1$11

and derives continuity of bilinear pseudo-differential operators on modulation spaces with exponential-type weights (Abdeljawad et al., 2019).

The periodic theory has a particularly transparent form. In (Toft et al., 2017), periodic Gevrey and Gelfand–Shilov classes are characterized both by Fourier coefficient estimates and by STFT estimates, and periodic modulation-space norms are identified with weighted $s>1$12-norms of Fourier coefficients. In the periodic setting this means that Gevrey regularity is equivalent to exponential Fourier coefficient decay of the form

$s>1$13

and, simultaneously, to exponential decay of the STFT in the frequency variable (Toft et al., 2017).

6. PDE motivation, extensions, and indirect frameworks

The original motivation is explicitly nonlinear PDE. The basic model given in (Reich, 2014) is that, to solve equations of the form

$s>1$14

one needs a function space stable under multiplication and under composition by the relevant nonlinearity. The algebra theorem covers polynomial and analytic nonlinearities, while the non-analytic superposition theorem covers a larger class of $s>1$15 nonlinearities defined through Fourier-measure conditions. The paper also records the homogeneous wave equation

$s>1$16

with solution

$s>1$17

as a linear model illustrating the modulation-space framework (Reich, 2014).

The 2015 development extends the theory to a weaker ultra-differentiable modulation scale. Using

$s>1$18

the paper defines $s>1$19 by

$s>1$20

It proves that $s>1$21 is an algebra and satisfies a non-analytic superposition theorem analogous to the Gevrey-modulation result. The paper also states the inclusion chain

$s>1$22

(Reich et al., 2015).

More recent work has produced related but distinct backgrounds. “Extended Gevrey regularity via weighted matrices” and “Gelfand-Shilov spaces for extended Gevrey regularity” study the sequence

$s>1$23

and its associated-function formalism. These papers do not define Gevrey-modulation spaces, but they provide a weight-matrix and time-frequency language for regularity scales lying between the union of Gevrey classes and $s>1$24, together with corresponding Gelfand–Shilov-type spaces and time-frequency representations (Teofanov et al., 2022, Teofanov et al., 17 Mar 2025). This suggests a broader family of modulation-space constructions beyond the classical Gevrey weight $s>1$25.

A different indirect development is the subdyadic modulation-space framework of (Vergara, 21 Nov 2025). That paper introduces a nonuniform Gabor-frame geometry adapted to dispersive phases and proves window and lattice independence, duality, almost diagonalization, and microlocal results. It is explicitly not a paper about Gevrey-modulation spaces: it remains in a polynomial-weight regime and “does not develop Gevrey, Gelfand–Shilov, or exponential/subexponential weighted modulation spaces.” Its relevance is structural rather than direct, since it isolates a geometry-plus-frame-plus-coefficient formalism that could plausibly be combined with Gevrey-type weights in future work (Vergara, 21 Nov 2025).

Taken together, these developments establish Gevrey-modulation spaces as a phase-space analogue of Gevrey regularity with three defining features: subexponential Fourier-side weights, stability under multiplication, and nontrivial superposition theory. Their distinctive role in the literature lies not merely in stronger smoothness, but in the coexistence of Gevrey-type frequency decay with nonlinear closure properties that are specifically designed for semilinear analysis (Reich, 2014).