Orthonormal Strichartz Estimates

• Orthonormal Strichartz estimates are mixed-norm bounds for densities of dispersive evolutions built from orthonormal families, providing a refined extension of classical estimates.
• They utilize a density-matrix and Schatten-class framework to capture sharper summability exponents and account for scaling, Sobolev regularity, and geometric interactions.
• These estimates have significant implications for kinetic transport and many-body quantum dynamics, linking operator duality to endpoint phenomena and spectral theory.

Orthonormal Strichartz estimates are mixed-norm bounds for densities of dispersive evolutions built from orthonormal families rather than a single initial datum. For the free Schrödinger flow, the central quantity is

$$\Big\|\sum_j \lambda_j\, |e^{it\Delta} f_j|^2\Big\|_{L_t^pL_x^q},$$

with $(f_j)_j$ orthonormal in $L^2(\mathbb R^d)$ or, more generally, in $\dot H^s(\mathbb R^d)$, and $\lambda=(\lambda_j)_j$ nonnegative. The subject originated in work of Frank, Lewin, Lieb, and Seiringer, which placed the theory in a density-matrix and Schatten-class framework, and it was substantially extended by Frank–Sabin and by Bez–Hong–Lee–Nakamura–Sawano to Sobolev regularity, sharp summability exponents, localized/global dichotomies, and endpoint phenomena (Frank et al., 2013, Bez et al., 2017).

1. Origin, formulation, and density-matrix viewpoint

The orthonormal theory generalizes the usual single-function Strichartz estimate in the same way that the Lieb–Thirring inequality generalizes the Sobolev inequality: the basic object is no longer $e^{it\Delta}f$, but the density

$$\rho_{\gamma(t)}(x)=\sum_j \lambda_j\,|e^{it\Delta}f_j(x)|^2,$$

where $\gamma_0=\sum_j \lambda_j |f_j\rangle\langle f_j|$ is the initial density operator. In the free Euclidean setting, the foundational inequality takes the form

$$\|\rho_{\gamma(t)}\|_{L_t^pL_x^q}\lesssim \|\gamma_0\|_{\mathcal C^\alpha},$$

or equivalently

$$\Big\|\sum_j \lambda_j\, |e^{it\Delta} f_j|^2\Big\|_{L_t^pL_x^q}\lesssim \|\lambda\|_{\ell^\alpha},$$

with $(f_j)_j$0 determined by the admissible exponents and the geometry of the problem (Frank et al., 2013).

The 2013 theory already identified the dual operator-valued formulation. If

$(f_j)_j$1

then orthonormal Strichartz bounds are equivalent to Schatten estimates for $(f_j)_j$2. This operator-theoretic viewpoint became a permanent feature of the subject, because it connects orthonormal Strichartz inequalities to density matrices, trace ideals, and many-body dynamics (Frank et al., 2013).

For Sobolev-regular initial data, the natural Hilbert structure changes. In the Euclidean Schrödinger setting, $(f_j)_j$3 is orthonormal in $(f_j)_j$4 when $(f_j)_j$5 is orthonormal in $(f_j)_j$6, and

$(f_j)_j$7

This replacement is not cosmetic: it changes the scaling line, the critical pair, and the optimal summability exponent (Bez et al., 2017).

2. Scaling, admissibility, and the sharp summability exponent

For the free Schrödinger propagator with Sobolev regularity $(f_j)_j$8, scaling dictates

$(f_j)_j$9

In the orthonormal problem, this scaling line interacts with an additional “orthonormal critical line”

$L^2(\mathbb R^d)$0

Their intersection gives the critical pair. For $L^2(\mathbb R^d)$1, this is

$L^2(\mathbb R^d)$2

whereas for general $L^2(\mathbb R^d)$3 it is

$L^2(\mathbb R^d)$4

This shift is a defining feature of the orthonormal theory: the critical case is not the classical endpoint of single-function Strichartz estimates (Bez et al., 2017).

The sharp candidate for the coefficient exponent is

$L^2(\mathbb R^d)$5

On the Schrödinger scaling line with $L^2(\mathbb R^d)$6, this gives $L^2(\mathbb R^d)$7. On the orthonormal critical line $L^2(\mathbb R^d)$8, $L^2(\mathbb R^d)$9, while on $\dot H^s(\mathbb R^d)$0, $\dot H^s(\mathbb R^d)$1 in the coordinate system used in the Euclidean Sobolev paper (Bez et al., 2017).

Two necessary conditions are basic. If an orthonormal Strichartz bound holds, then necessarily

$\dot H^s(\mathbb R^d)$2

The first is proved by a wave-packet tube construction using frequency-separated initial data; the second is proved by a time-shifted orthonormal system $\dot H^s(\mathbb R^d)$3 with suitable $\dot H^s(\mathbb R^d)$4 (Bez et al., 2017).

A common misconception is that orthogonality only improves constants. In fact, it changes the summability problem itself: the sharp exponent $\dot H^s(\mathbb R^d)$5 becomes part of the theorem, and its optimal value depends on the location of $\dot H^s(\mathbb R^d)$6 relative to the orthonormal critical line (Bez et al., 2017).

3. Euclidean Schrödinger theory with Sobolev regularity

The central Euclidean result for regular data is the strong-type theorem in the subcritical region. If $\dot H^s(\mathbb R^d)$7 lies in the interior of $\dot H^s(\mathbb R^d)$8, $\dot H^s(\mathbb R^d)$9, and $\lambda=(\lambda_j)_j$0, then

$\lambda=(\lambda_j)_j$1

for all orthonormal systems in $\lambda=(\lambda_j)_j$2. This is sharp in the sense that the inequality fails for $\lambda=(\lambda_j)_j$3 (Bez et al., 2017).

In the near-critical region $\lambda=(\lambda_j)_j$4, the picture changes. For $\lambda=(\lambda_j)_j$5 and $\lambda=(\lambda_j)_j$6 in the interior of $\lambda=(\lambda_j)_j$7, one has strong-type bounds for every $\lambda=(\lambda_j)_j$8, and failure for $\lambda=(\lambda_j)_j$9. Thus the barrier $e^{it\Delta}f$0 is structural, not technical (Bez et al., 2017).

The decisive distinction from the classical theory appears on the orthonormal critical line $e^{it\Delta}f$1. For $e^{it\Delta}f$2, if $e^{it\Delta}f$3 is a compactly supported Fourier multiplier, then the frequency-localized bound

$e^{it\Delta}f$4

holds on $e^{it\Delta}f$5, and is sharp against $e^{it\Delta}f$6. Without localization, however, the strong-type inequality at $e^{it\Delta}f$7 fails; more precisely, even the weak Lorentz-space substitute fails for all $e^{it\Delta}f$8 on $e^{it\Delta}f$9 (Bez et al., 2017). The paper states the conceptual reason explicitly: localization suppresses long-range time–space interactions, while globally orthonormal densities can concentrate across scales in a way that defeats summability at $$\rho_{\gamma(t)}(x)=\sum_j \lambda_j\,|e^{it\Delta}f_j(x)|^2,$$0 (Bez et al., 2017).

Weak-type estimates partially fill the gap. Across the subcritical region, one has restricted weak-type

$$\rho_{\gamma(t)}(x)=\sum_j \lambda_j\,|e^{it\Delta}f_j(x)|^2,$$1

with $$\rho_{\gamma(t)}(x)=\sum_j \lambda_j\,|e^{it\Delta}f_j(x)|^2,$$2. On the segment $$\rho_{\gamma(t)}(x)=\sum_j \lambda_j\,|e^{it\Delta}f_j(x)|^2,$$3, one also has

$$\rho_{\gamma(t)}(x)=\sum_j \lambda_j\,|e^{it\Delta}f_j(x)|^2,$$4

At the endpoint $$\rho_{\gamma(t)}(x)=\sum_j \lambda_j\,|e^{it\Delta}f_j(x)|^2,$$5 for $$\rho_{\gamma(t)}(x)=\sum_j \lambda_j\,|e^{it\Delta}f_j(x)|^2,$$6, the FLLS conjecture predicts restricted weak-type with $$\rho_{\gamma(t)}(x)=\sum_j \lambda_j\,|e^{it\Delta}f_j(x)|^2,$$7; it is false in $$\rho_{\gamma(t)}(x)=\sum_j \lambda_j\,|e^{it\Delta}f_j(x)|^2,$$8, while for $$\rho_{\gamma(t)}(x)=\sum_j \lambda_j\,|e^{it\Delta}f_j(x)|^2,$$9 it remains open (Bez et al., 2017).

A later development settled the critical summability exponent in the interior of the region denoted $\gamma_0=\sum_j \lambda_j |f_j\rangle\langle f_j|$0 in that paper’s notation. Specifically, for $\gamma_0=\sum_j \lambda_j |f_j\rangle\langle f_j|$1 and $\gamma_0=\sum_j \lambda_j |f_j\rangle\langle f_j|$2, one has

$\gamma_0=\sum_j \lambda_j |f_j\rangle\langle f_j|$3

with $\gamma_0=\sum_j \lambda_j |f_j\rangle\langle f_j|$4, obtained from global restricted weak-type at $\gamma_0=\sum_j \lambda_j |f_j\rangle\langle f_j|$5, real interpolation, and the crucial inequality $\gamma_0=\sum_j \lambda_j |f_j\rangle\langle f_j|$6 in that region (Feng et al., 20 Jul 2025). This uses a different symbol convention for the time exponent, but it resolves the interior critical-summability problem left open by the earlier Sobolev theory (Feng et al., 20 Jul 2025).

4. Duality, interpolation, and obstruction mechanisms

The duality principle is one of the structural pillars of the subject. In the Euclidean Schrödinger setting, if $\gamma_0=\sum_j \lambda_j |f_j\rangle\langle f_j|$7 and $\gamma_0=\sum_j \lambda_j |f_j\rangle\langle f_j|$8 is a multiplication operator, then

$\gamma_0=\sum_j \lambda_j |f_j\rangle\langle f_j|$9

is equivalent to

$$\|\rho_{\gamma(t)}\|_{L_t^pL_x^q}\lesssim \|\gamma_0\|_{\mathcal C^\alpha},$$0

The same operator-theoretic equivalence underlies abstract measure-space formulations, Dunkl theory, compact manifolds, and perturbative settings (Bez et al., 2017).

On the proof side, the Euclidean Sobolev theory combines bilinear real interpolation in the style of Keel–Tao on dyadic time pieces of $$\|\rho_{\gamma(t)}\|_{L_t^pL_x^q}\lesssim \|\gamma_0\|_{\mathcal C^\alpha},$$1, Littlewood–Paley decomposition, a Bourgain trick for vector-valued gluing, real and complex interpolation in mixed-norm spaces, and Lorentz-space Hardy–Littlewood–Sobolev inequalities (Bez et al., 2017). For wave, Klein–Gordon, and fractional Schrödinger equations, weighted oscillatory integral estimates become central; these deliver the optimal decay exponents needed to run the Frank–Sabin Schatten $$\|\rho_{\gamma(t)}\|_{L_t^pL_x^q}\lesssim \|\gamma_0\|_{\mathcal C^\alpha},$$2 machinery beyond the diagonal cases (Bez et al., 2019).

The obstruction mechanisms are equally explicit. The condition $$\|\rho_{\gamma(t)}\|_{L_t^pL_x^q}\lesssim \|\gamma_0\|_{\mathcal C^\alpha},$$3 comes from wave-packet tubes: choose disjointly supported frequency packets so that each evolution is essentially supported on a tube $$\|\rho_{\gamma(t)}\|_{L_t^pL_x^q}\lesssim \|\gamma_0\|_{\mathcal C^\alpha},$$4, sum the characteristic functions, and compare the left- and right-hand sides. The condition $$\|\rho_{\gamma(t)}\|_{L_t^pL_x^q}\lesssim \|\gamma_0\|_{\mathcal C^\alpha},$$5 comes from time-translated orthonormal systems $$\|\rho_{\gamma(t)}\|_{L_t^pL_x^q}\lesssim \|\gamma_0\|_{\mathcal C^\alpha},$$6, for which one obtains

$$\|\rho_{\gamma(t)}\|_{L_t^pL_x^q}\lesssim \|\gamma_0\|_{\mathcal C^\alpha},$$7

(Bez et al., 2017).

A further obstruction emerges through semiclassical limits. If an orthonormal Strichartz estimate holds, then a corresponding weighted velocity-average estimate for kinetic transport follows. This principle is used both positively, to derive transport estimates, and negatively, to show endpoint failures (Bez et al., 2017). On scattering manifolds, the same semiclassical mechanism shows that bounded invariant sets for the Hamiltonian flow, and in particular periodic stable geodesics, break the sharp orthonormal Strichartz estimates (Hoshiya, 5 Apr 2025).

An important abstract development is the Keel–Tao type theorem for orthonormal Strichartz estimates: dispersive $$\|\rho_{\gamma(t)}\|_{L_t^pL_x^q}\lesssim \|\gamma_0\|_{\mathcal C^\alpha},$$8 bounds for strongly continuous unitary groups imply orthonormal Strichartz inequalities for the corresponding $$\|\rho_{\gamma(t)}\|_{L_t^pL_x^q}\lesssim \|\gamma_0\|_{\mathcal C^\alpha},$$9-admissible pairs. This applies to settings such as unbounded electromagnetic potentials, $$\Big\|\sum_j \lambda_j\, |e^{it\Delta} f_j|^2\Big\|_{L_t^pL_x^q}\lesssim \|\lambda\|_{\ell^\alpha},$$0-generalized Laguerre operators, and scaling-critical magnetic Hamiltonians (Hoshiya, 2024).

5. Generalizations across equations, geometries, and operators

The theory no longer belongs exclusively to the free Schrödinger equation on $$\Big\|\sum_j \lambda_j\, |e^{it\Delta} f_j|^2\Big\|_{L_t^pL_x^q}\lesssim \|\lambda\|_{\ell^\alpha},$$1. Weighted oscillatory integral methods yielded orthonormal Strichartz estimates for the wave, Klein–Gordon, and fractional Schrödinger equations, with sharp $$\Big\|\sum_j \lambda_j\, |e^{it\Delta} f_j|^2\Big\|_{L_t^pL_x^q}\lesssim \|\lambda\|_{\ell^\alpha},$$2 on substantial portions of the sharp admissible lines and $$\Big\|\sum_j \lambda_j\, |e^{it\Delta} f_j|^2\Big\|_{L_t^pL_x^q}\lesssim \|\lambda\|_{\ell^\alpha},$$3 further along the lines, together with applications to weighted velocity averaging and Hartree-type systems (Bez et al., 2019). A later note on wave equations isolates maximal-in-space boundary cases and formulates a conjectural sharp picture for the remaining open wave regimes (Bez et al., 2023).

On abstract measure spaces, a non-negative self-adjoint operator $$\Big\|\sum_j \lambda_j\, |e^{it\Delta} f_j|^2\Big\|_{L_t^pL_x^q}\lesssim \|\lambda\|_{\ell^\alpha},$$4 with kernel decay

$$\Big\|\sum_j \lambda_j\, |e^{it\Delta} f_j|^2\Big\|_{L_t^pL_x^q}\lesssim \|\lambda\|_{\ell^\alpha},$$5

gives both single-function and orthonormal Strichartz estimates for $$\Big\|\sum_j \lambda_j\, |e^{it\Delta} f_j|^2\Big\|_{L_t^pL_x^q}\lesssim \|\lambda\|_{\ell^\alpha},$$6, as well as for frequency-localized semigroups $$\Big\|\sum_j \lambda_j\, |e^{it\Delta} f_j|^2\Big\|_{L_t^pL_x^q}\lesssim \|\lambda\|_{\ell^\alpha},$$7. This formulation unifies Euclidean space, Hermite and Laguerre settings, twisted Laplacians, and several other dispersive models (Feng et al., 2024).

Perturbative theories now cover Schrödinger operators with potentials. For time-independent short-range, inverse-square, and magnetic potentials, orthonormal Strichartz estimates are transferred from the free flow by Kato smooth perturbation theory. This yields global-in-time bounds for $$\Big\|\sum_j \lambda_j\, |e^{it\Delta} f_j|^2\Big\|_{L_t^pL_x^q}\lesssim \|\lambda\|_{\ell^\alpha},$$8 in the same admissible regimes as the free case, plus refined Besov-space versions, and applications to infinitely many fermions with electromagnetic potentials (Hoshiya, 2023). For repulsive Hamiltonians, a Keel–Tao type orthonormal theory is combined with uniform resolvent estimates with logarithmic decaying weight functions to obtain new global-in-time estimates and smoothing results (Hoshiya, 2024).

On compact manifolds, the sharp-line theory for fractional Schrödinger, wave, and Klein–Gordon equations is frequency localized and carries $$\Big\|\sum_j \lambda_j\, |e^{it\Delta} f_j|^2\Big\|_{L_t^pL_x^q}\lesssim \|\lambda\|_{\ell^\alpha},$$9 losses matching the corresponding single-function derivative losses. On the sphere these bounds can be saturated, while on the flat torus they can improve through decoupling for non-smooth hypersurfaces (Wang et al., 11 Mar 2025). A subsequent paper extends the compact-manifold theory to the non-sharp admissible region, using a Lieb–Sobolev inequality derived from a Cwikel estimate and an alternative globalization method based on localized weak Lorentz estimates (Ji et al., 10 May 2026).

Periodic and partially periodic settings display their own phenomena. On the flat torus, mixed-norm orthonormal Strichartz estimates on the lower region $(f_j)_j$00 match the Euclidean exponent $(f_j)_j$01 without $(f_j)_j$02-loss after localization to intervals of length $(f_j)_j$03; the one-dimensional endpoint gives a sharp $(f_j)_j$04 inequality with $(f_j)_j$05 (Nakamura, 2018). More recently, fractional Schrödinger estimates on $(f_j)_j$06 and on waveguide manifolds $(f_j)_j$07 were established, with explicit $(f_j)_j$08-losses and $(f_j)_j$09, together with improved $(f_j)_j$10 decoupling and Hartree applications (Bhimani et al., 22 Jul 2025). Refined torus estimates in mixed Lebesgue spaces with partial regularity extend Nakamura’s orthonormal theory and lead to well-posedness for Hartree equations in Schatten spaces (Bhimani et al., 28 Jan 2026).

The Dunkl and generalized Laguerre settings replace the Euclidean dimension by an effective dimension. For the Dunkl–Schrödinger equation, the effective dimension is $(f_j)_j$11, and the scaling relation becomes

$(f_j)_j$12

The same orthonormal geometry reappears, with frequency-localized, restricted weak-type, and global Sobolev estimates parallel to the Euclidean case (Feng et al., 5 Jun 2025). Earlier work established orthonormal Strichartz estimates for $(f_j)_j$13-generalized Laguerre operators and transferred them to Dunkl operators through explicit kernel relations (Mondal et al., 2022).

6. Applications, endpoint phenomena, and open directions

A primary application is kinetic transport. The semiclassical principle in the Euclidean Sobolev theory implies weighted velocity-averaging estimates of the form

$(f_j)_j$14

for $(f_j)_j$15 in the subcritical region, and the newer critical-summability results extend this to the $(f_j)_j$16 interior in the notation of the later paper (Bez et al., 2017, Feng et al., 20 Jul 2025). On nontrapping scattering manifolds, global-in-time orthonormal Strichartz estimates imply global transport Strichartz bounds for kinetic equations in the semiclassical limit and feed into small-data scattering for the cutoff Boltzmann equation (Hoshiya, 5 Apr 2025).

Many-body quantum dynamics is another persistent theme. Orthonormal Strichartz estimates control densities $(f_j)_j$17 and furnish bounds in Schatten classes, which underpin Hartree dynamics with infinitely many fermions. This role is explicit in Euclidean, torus, abstract-measure, compact-manifold, and perturbative-potential settings (Bez et al., 2017, Nakamura, 2018). The torus and waveguide fractional theories also use orthonormal Strichartz estimates to treat Hartree equations with non-trace-class initial data (Bhimani et al., 22 Jul 2025).

Endpoint behavior remains the main unresolved issue. In the Euclidean Schrödinger theory, the global $(f_j)_j$18 bound fails on the orthonormal critical line, while the restricted weak-type conjecture at the critical point $(f_j)_j$19 is still open for $(f_j)_j$20 (Bez et al., 2017). In the notation of the later critical-summability paper, strong-type at $(f_j)_j$21 is now known in the interior of $(f_j)_j$22, but the boundary segment $(f_j)_j$23 still has only restricted weak-type at $(f_j)_j$24, and upgrading $(f_j)_j$25 to strong type remains open (Feng et al., 20 Jul 2025). On the circle, renormalization of the density improves the $(f_j)_j$26 estimate from $(f_j)_j$27 to $(f_j)_j$28, and improves the $(f_j)_j$29 range as well, but the conjectured optimal renormalized $(f_j)_j$30 range down to $(f_j)_j$31 is unresolved (Hadama et al., 31 Mar 2026).

A broader lesson is that orthonormal Strichartz estimates are not a uniform extension of classical Strichartz theory. They have their own critical lines, their own sharp coefficient exponents, and their own geometry-sensitive failures. Frequency localization can restore critical summability that fails globally; periodic geometry introduces unavoidable $(f_j)_j$32-losses; stable periodic geodesics obstruct sharp global estimates; and effective dimensions in Dunkl-type settings modify every scaling law (Bez et al., 2017, Hoshiya, 5 Apr 2025). This suggests that future progress will continue to depend on combining operator-theoretic duality with geometry-specific microlocal and interpolation methods.