Skip to content
BEST·BOOKS
+ MENU
Books Authors Tags Endorsers Sources
← Back to Nonlinear Dispersive Equations: Local and Global Analysis

AI Study Notebook AI-generated

Study Guide: Nonlinear Dispersive Equations: Local and Global Analysis

Terence Tao

By Best Books

This AI-generated study guide is a reading aid. The source-backed recommendation record and evidence for this book live on the book page.

Key points Not available Flashcards Not available
On this page

Nonlinear Dispersive Equations: Local and Global Analysis — Chapter-by-Chapter Outline

Author: Terence Tao First published: 2006 Edition covered: First and only edition, CBMS Regional Conference Series in Mathematics, Volume 106, American Mathematical Society, Providence, RI (2006). The book originated as lecture notes for an NSF-CBMS regional conference held at New Mexico State University in June 2005. No revised edition has been published; a freely available online preprint of Chapter 1 predates the printed version.

Central thesis

Nonlinear dispersive equations — including the nonlinear Schrödinger equation (NLS), the nonlinear wave equation (NLW), the Korteweg–de Vries equation (KdV), and the wave maps equation (WM) — share a common analytic architecture. Understanding that architecture, rather than treating each equation in isolation, is the goal of the book. The architecture rests on three pillars: dispersion (high-frequency components propagate at different speeds, causing solutions to spread out), nonlinearity (the spreading can be countered or amplified depending on the sign and power of the nonlinear term), and criticality (the ratio between nonlinear and dispersive effects depends on the scale of the solution, and whether the nonlinearity is sub-, super-, or critical relative to scaling determines which techniques apply).

Tao argues that the central problem in the field — constructing solutions that exist globally in time for large initial data — requires a layered strategy: start with the simpler ODE setting to absorb the tools cleanly, move to linear PDE to get dispersion estimates, then build nonlinear local theory via fixed-point arguments, then extend globally using conservation laws, monotonicity, and increasingly sophisticated induction-on-energy or concentration-compactness methods. Each layer of the book adds one new difficulty while relying on the toolkit of all previous layers.

The book's governing question is:

Given an initial condition for a nonlinear dispersive PDE, does the solution exist for all time, and if so, how does it behave as time tends to infinity?

Chapter 1 — Ordinary Differential Equations

Central question

What fundamental analytic tools — existence, uniqueness, conservation laws, monotonicity, and integrability — can be illustrated in the simpler setting of ordinary differential equations, before being deployed in the infinite-dimensional PDE context?

Main argument

General theory of ODEs as a prototype. The chapter begins with the Picard–Lindelöf theorem and its proof via contraction mapping on the space of continuous functions. This is not merely a warm-up: the fixed-point argument reappears verbatim in Chapter 3 to prove local well-posedness for NLS and NLW. Tao is explicit that the ODE proofs are templates, and he writes them in a form that generalizes. The notion of a lifespan of a solution is introduced: the maximal interval of existence, with blowup occurring when a suitable norm diverges.

Gronwall's inequality. Section 1.2 develops Gronwall's inequality in its integral form: if $y(t) \leq A + B\int_0^t y(s)\,ds$ then $y(t) \leq A e^{Bt}$. This bound controls how quickly solutions to perturbed equations diverge from each other. The ODE version feeds directly into perturbative stability arguments for PDE in later chapters.

Bootstrap and continuity arguments. Section 1.3 presents the bootstrap principle, perhaps the single most-used technique in the book. The idea is to assume a stronger-than-necessary estimate, derive it from itself under a smallness or continuity assumption, and then close the argument. Tao gives it a precise formulation: if a property $P(t)$ is open, closed, and holds at $t=0$, then it holds for all $t$ in the connected interval. This is the qualitative version of the contraction argument.

Noether's theorem. Section 1.4 translates Noether's theorem into the ODE setting: if an ODE has a one-parameter family of symmetries generated by a vector field $V$, then the quantity $Q = \langle p, V(q)\rangle$ (the Noether charge) is conserved along trajectories. Concretely: time-translation symmetry implies energy conservation; spatial-translation symmetry implies momentum conservation; phase-rotation symmetry for complex-valued ODEs implies charge (mass) conservation. These are exactly the quantities that recur in NLS and NLW.

Monotonicity formulae. Section 1.5 introduces the idea of Lyapunov functions and more generally quantities that are monotone (non-increasing or non-decreasing) rather than exactly conserved. In the ODE setting this is illustrated with the virial identity, a computation showing that $\frac{d^2}{dt^2}|q|^2 = 2|p|^2 + 2\langle q, F(q)\rangle$. When the right side is sign-definite, one can prove that solutions cannot remain localized forever — a precursor to the Morawetz estimates of Chapter 5.

Linear and semilinear ODEs. Section 1.6 treats the Duhamel formula $u(t) = e^{tA}u(0) + \int0^t e^{(t-s)A} F(u(s))\,ds$ for the ODE system $\partialt u = Au + F(u)$. This representation of the solution as a free evolution plus a retarded nonlinear forcing is the Duhamel formula, which appears identically in every subsequent chapter for PDE.

Completely integrable systems. Section 1.7 introduces ODEs with sufficiently many conserved quantities to be solvable by quadrature — Liouville–Arnold integrability. The ODE analogs of the KdV and NLS equations (the harmonic oscillator and Toda lattice) are discussed. This section frames the completely integrable PDE of Chapter 4 against a finite-dimensional background.

Key ideas

  • The fixed-point (Picard iteration) proof of local existence is the universal template for PDE well-posedness.
  • Gronwall's inequality quantifies continuity of solutions with respect to data and perturbations.
  • The bootstrap principle formalizes how one can assume a bound, derive it from itself, and close an argument.
  • Noether's theorem converts each continuous symmetry into a conserved charge; these charges (energy, momentum, mass) become the primary global tools in later chapters.
  • Monotonicity formulae, not just conservation laws, provide one-sided control that suffices for scattering and blowup results.
  • The Duhamel formula reduces the nonlinear problem to a fixed-point problem for an integral equation; this reduction is used without comment in every subsequent chapter.
  • Completely integrable ODE systems forecast the special structure of KdV and its infinite tower of conserved quantities.

Key takeaway

Chapter 1 builds the complete toolbox — fixed-point arguments, Gronwall bounds, bootstrap, Noether charges, monotonicity, Duhamel — in the simpler finite-dimensional ODE setting so that every technique appears first without the additional complexity of function spaces and Fourier analysis.

Chapter 2 — Constant Coefficient Linear Dispersive Equations

Central question

How do the solutions of linear dispersive PDE — the Schrödinger equation $i\partialt u + \Delta u = 0$ and the wave equation $\partial{tt} u - \Delta u = 0$ — propagate, disperse, and concentrate, and what are the quantitative estimates that encode this behavior?

Main argument

The Fourier transform and the dispersion relation. Section 2.1 sets up the Fourier-side representation. A linear constant-coefficient evolution equation has the form $\partialt u = P(D)u$ where $P(D)$ is a polynomial in the gradient. The solution is $\hat u(t,\xi) = e^{P(i\xi)t}\hat u0(\xi)$. The dispersion relation $\tau = h(\xi)$ encodes the frequency-dependence of phase velocity; the equation is dispersive when the group velocity $\nabla_\xi h(\xi)$ is not constant, so different frequencies travel at different speeds and localized packets spread out. The Schrödinger dispersion relation is $\tau = |\xi|^2$ (quadratic, group velocity $2\xi$); the wave equation has $\tau = |\xi|$ (linear, constant group speed, but direction-dependent).

Fundamental solution and stationary phase. Section 2.2 computes the fundamental solution (the kernel $K(t,x)$ such that $u = Ku_0$) via stationary phase. For the Schrödinger equation in $\mathbb{R}^d$, $K(t,x) = (4\pi i t)^{-d/2} e^{i|x|^2/4t}$, giving the *dispersive decay estimate** $|u(t)|{L^\infty} \lesssim |t|^{-d/2}|u0|_{L^1}$. Solutions spread at rate $|t|^{-d/2}$, and the point-wise $L^\infty$ norm decays. This is the quantitative content of "dispersive."

Strichartz estimates. Section 2.3 is the technical core. By interpolating between the $L^\infty$ dispersive decay and the $L^2$ conservation of norm (Plancherel), one obtains the Strichartz estimates: $|e^{it\Delta}u0|{L^pt L^qx} \lesssim |u0|{L^2}$ for a range of exponents $(p,q)$ satisfying the admissibility condition $\frac{2}{p} + \frac{d}{q} = \frac{d}{2}$. The endpoint was proved by Keel and Tao (1998). Strichartz estimates convert $L^2$ control of initial data into space-time $L^p L^q$ control of the solution — the key input for the contraction argument of Chapter 3. Retarded Strichartz estimates for the inhomogeneous problem $i\partial_t u + \Delta u = F$ are also established.

Conservation laws for the Schrödinger equation. Section 2.4 uses Noether's theorem (from Chapter 1) to derive the mass conservation law $\partial_t |\hat u|^2 + \nabla \cdot J = 0$, the momentum density, and the energy. These identities are computed explicitly as integral identities rather than formal Noether outputs, to make them usable in the PDE setting.

The wave equation stress-energy tensor. Section 2.5 carries out the same programme for the wave equation. The stress-energy tensor $T^{\mu\nu}$ is constructed; conservation of $T^{00}$ is energy conservation; conservation of $T^{0i}$ is momentum conservation. The tensor formalism keeps the computations covariant and generalizes naturally to the geometric (wave maps) setting of Chapter 6.

$X^{s,b}$ spaces (Bourgain spaces). Section 2.6 introduces the $X^{s,b}$ spaces (also called Bourgain spaces): the space of functions whose Fourier transform in both space and time is controlled by $\langle\xi\rangle^s \langle\tau - h(\xi)\rangle^b$. These spaces measure how well a function is concentrated near the characteristic surface $\tau = h(\xi)$ of the linear equation. They capture the transference of Strichartz-type estimates to the bilinear setting and are essential for the low-regularity well-posedness theory of KdV in Chapter 4.

Key ideas

  • The dispersion relation $\tau = h(\xi)$ is the fundamental object; the equation is dispersive iff the group velocity $\nabla h$ is non-constant.
  • The fundamental solution decays pointwise at rate $|t|^{-d/2}$, a quantitative expression of dispersion.
  • Strichartz estimates convert this decay into space-time mixed-norm bounds; admissible pairs $(p,q)$ are those satisfying the scaling condition $\frac{2}{p}+\frac{d}{q}=\frac{d}{2}$.
  • Conservation laws (mass, momentum, energy) for both the Schrödinger and wave equations are derived from symmetries via Noether's theorem.
  • The stress-energy tensor for the wave equation provides a systematic way to encode all conservation laws in one covariant object.
  • $X^{s,b}$ spaces allow one to measure proximity to the free evolution and are the natural function spaces for bilinear estimates.

Key takeaway

Chapter 2 establishes the quantitative dispersive toolkit — Strichartz estimates, the fundamental solution, conservation laws, and Bourgain spaces — that underlies every well-posedness argument in the book.

Chapter 3 — Semilinear Dispersive Equations

Central question

How does one construct local and global solutions to nonlinear Schrödinger (NLS) and nonlinear wave (NLW) equations, quantify their stability, and understand their long-time behavior including scattering?

Main argument

Scaling and criticality. Section 3.1 is the conceptual keystone of the whole book. For the NLS $i\partialt u + \Delta u = \lambda |u|^{p-1}u$ in $\mathbb{R}^d$, the scaling symmetry $u(t,x) \mapsto \lambda^{2/(p-1)} u(\lambda^2 t, \lambda x)$ maps solutions to solutions. The critical Sobolev exponent is $sc = \frac{d}{2} - \frac{2}{p-1}$, the unique Sobolev index that is invariant under the scaling. When the regularity of the data is above $sc$ (subcritical), the local existence time scales favorably; when it equals $sc$ (critical), the problem is scale-invariant and much harder; when it is below $s_c$ (supercritical), perturbative methods fail entirely. This trichotomy organizes the chapter and the rest of the book.

What is a solution? Section 3.2 addresses the definition of solution. Tao distinguishes classical, strong, mild, and distributional solutions, and argues that the mild (Duhamel) formulation $u(t) = e^{it\Delta}u0 - i\lambda\int0^t e^{i(t-s)\Delta}|u(s)|^{p-1}u(s)\,ds$ is the natural setting for well-posedness proofs. The question of persistence of regularity — whether a mild solution with rough data can be upgraded to a classical solution given smoother data — is also discussed.

Local existence theory. Section 3.3 applies the contraction mapping principle in Strichartz spaces to prove local well-posedness for NLS and NLW in the subcritical and critical regimes. The argument: define the iteration map $\Phi[u](t) = e^{it\Delta}u0 - i\lambda\int0^t e^{i(t-s)\Delta}N(u(s))\,ds$; show $\Phi$ maps a ball to itself (using Strichartz and Sobolev inequalities); show $\Phi$ is a contraction on that ball. The lifespan depends on the norm of initial data in the subcritical case, but only on the norm in the critical case (yielding small-data global existence). Energy methods (multiplying the equation by $\bar u$ and integrating) give an alternative route for lower regularities.

Conservation laws and global existence. Section 3.4 uses the mass conservation law $|u(t)|{L^2}^2 = |u0|{L^2}^2$ and the energy conservation $H[u] = \frac{1}{2}|\nabla u|{L^2}^2 + \frac{\lambda}{p+1}|u|{L^{p+1}}^{p+1}$ to extend solutions beyond the local existence time. For the defocusing NLS ($\lambda > 0$), the energy is coercive: $|\nabla u|{L^2}$ is uniformly bounded, and local solutions can be iterated indefinitely to give global ones. For the focusing NLS ($\lambda < 0$), the negative potential energy can swallow the kinetic energy if the initial data is too large, causing blowup. The interplay between focusing/defocusing and criticality is the fundamental tension in the subject.

Decay estimates. Section 3.5 establishes pointwise decay for small or scattering solutions. Using the pseudo-conformal identity (a space-time virial identity) and the decay of the linear propagator, Tao shows that solutions in the scattering regime satisfy $|u(t)|_{L^\infty} = O(t^{-d/2})$ — the same rate as the linear equation. These estimates require control of the nonlinear solution in spaces that interpolate Strichartz norms.

Scattering theory. Section 3.6 makes precise the idea that a global solution "looks linear" at late times. A solution scatters forward if there exists $u+\in L^2$ such that $|u(t) - e^{it\Delta}u+|_{L^2} \to 0$ as $t\to\infty$. Tao proves that small-data global solutions to subcritical defocusing NLS scatter, and discusses the connection to the wave operators. The defocusing cubic NLS in 3D is worked out explicitly as a model case.

Stability theory. Section 3.7 proves that if $\tilde u$ is an approximate solution of NLS (satisfying the equation up to an error $e$) and the error is small in a Strichartz norm, then there is a nearby exact solution. This perturbation lemma is an essential ingredient in the induction-on-energy arguments of Chapter 5.

Illposedness results. Section 3.8 shows that well-posedness fails below the critical regularity $s_c$ in several senses: the data-to-solution map fails to be uniformly continuous (norm inflation), or the solution map fails to be $C^2$. The proofs use explicit counterexamples (oscillating initial data) to show that the Picard iteration diverges in the appropriate function space.

Almost conservation laws. Section 3.9 introduces the I-method (or method of almost conservation laws), developed by Colliander, Keel, Staffilani, Takaoka, and Tao. When the natural conservation law (energy) does not control the Sobolev norm at which one wants to prove global well-posedness, one introduces a modified energy $E[Iu]$ where $I$ is a Fourier multiplier that smooths out high frequencies. The modified energy satisfies $\frac{d}{dt}E[Iu] = O(N^{-\varepsilon})$ (where $N$ is a frequency cutoff), so it is "almost" conserved. Iterating over time intervals of length $\sim N^\varepsilon$ gives global solutions below the energy regularity.

Key ideas

  • The critical Sobolev exponent $s_c = d/2 - 2/(p-1)$ is the dividing line between perturbative and non-perturbative regimes; the entire structure of the subject is organized around it.
  • Local well-posedness follows from a Strichartz-space contraction argument; the lifespan depends on $|u0|{H^s}$ subcritically, only on profile criticality.
  • Mass and energy conservation extend subcritical defocusing solutions to all time; focusing solutions may blowup.
  • The perturbation lemma (stability theory) is an algebraic abstraction of the contraction argument; it is the key ingredient in global large-data results.
  • Scattering (the solution looks like a free solution at late times) holds for small or defocusing subcritical solutions and requires decay of the nonlinear term in Strichartz norms.
  • Illposedness below $s_c$ shows the theory is sharp: the method cannot be pushed further without new ideas.
  • The I-method extends global theory below the energy threshold by using a frequency-truncated modified energy that is almost (but not exactly) conserved.

Key takeaway

Chapter 3 builds the complete nonlinear local and global theory for the two canonical semilinear equations (NLS and NLW), introducing the full hierarchy from scaling analysis through local existence, conservation laws, scattering, stability, illposedness, and the I-method — the framework within which all remaining chapters operate.

Chapter 4 — The Korteweg–de Vries Equation

Central question

How does one prove well-posedness for the KdV equation $\partialt u + \partial{xxx} u + u\partial_x u = 0$, which is more degenerate in its bilinear structure than NLS/NLW, and how does its complete integrability — infinite conservation laws, solitons, inverse scattering — affect the analysis?

Main argument

Existence theory and $X^{s,b}$ spaces. Section 4.1 proves local well-posedness for KdV in $H^s(\mathbb{R})$ for $s \geq -3/4$ using the Bourgain $X^{s,b}$ spaces introduced in Chapter 2. The key challenge is that the bilinear product $u\partial_x u$ involves a derivative and a product, and standard Strichartz estimates do not directly handle this. The strategy: write the Duhamel formula in $X^{s,b}$, use multilinear estimates in $X^{s,b}$ to close the contraction argument. The threshold $s = -3/4$ was established by Kenig, Ponce, and Vega; below it, bilinear estimates fail due to logarithmic divergences from high-high frequency interactions.

Correction terms and the modified KdV (mKdV). Section 4.2 introduces the Miura transform $u = vx + v^2$ which maps solutions of the modified KdV $\partialt v + \partial{xxx} v \pm 6v^2\partialx v = 0$ to solutions of KdV. This gauge-type transform reduces questions about KdV to questions about mKdV in certain regimes. For mKdV, additional conserved quantities are accessible (the $H^1$ norm is conserved exactly), allowing global results that can then be transferred back to KdV.

Symplectic non-squeezing. Section 4.3 applies Gromov's non-squeezing theorem to the infinite-dimensional setting of the KdV flow. The KdV flow on $L^2$ (viewed as an infinite-dimensional symplectic manifold) cannot map a ball of radius $r$ into a cylinder of radius $R < r$ in any coordinate hyperplane. Tao (2006) adapted Kuksin's infinite-dimensional symplectic geometry to prove this for KdV. The result is a qualitative statement about the global flow that goes beyond energy methods.

The Benjamin–Ono equation and gauge transforms. Section 4.4 treats the Benjamin–Ono equation $\partialt u + H\partial{xx} u + u\partialx u = 0$, where $H$ is the Hilbert transform. Benjamin–Ono is less dispersive than KdV ($|\xi|^2$ dispersion vs $|\xi|^3$), making the contraction argument in $X^{s,b}$ fail at low regularity. Tao's gauge transform for Benjamin–Ono maps the equation to a new form in which the nonlinear term has an additional $\partialx$ factored out, restoring enough structure for a contraction argument at regularity $H^0 = L^2$.

Key ideas

  • The $X^{s,b}$ (Bourgain space) contraction argument applies to KdV when supplemented with bilinear refinements; the critical regularity is $H^{-3/4}$.
  • Complete integrability of KdV provides infinitely many conserved quantities (one for each $H^k$), allowing global well-posedness for any fixed regularity.
  • The Miura transform $u = v_x + v^2$ links KdV to mKdV; exploiting the additional structure of mKdV yields sharper results for KdV.
  • Symplectic non-squeezing is a global geometric property of the flow that cannot be seen from local or energy arguments alone.
  • The Benjamin–Ono equation requires a gauge transform to compensate for its weaker dispersion; Tao's gauge transform is a nonlinear change of dependent variable that restores the contraction property.
  • KdV solitons $u(t,x) = 12c^2 \text{sech}^2(c(x - 4c^2 t))$ are exact solutions that travel without dispersing; the inverse scattering method produces all asymptotic states.

Key takeaway

Chapter 4 uses KdV to show that completely integrable equations have special structure — infinitely many conservation laws, solitons, gauge transforms — that must be exploited to prove sharp well-posedness, and that geometric methods (symplectic non-squeezing) can capture global properties of the flow beyond what energy or Strichartz methods give.

Chapter 5 — Energy-Critical Semilinear Dispersive Equations

Central question

How does one prove global existence and scattering for the energy-critical defocusing NLW and NLS (where the nonlinearity is exactly at the critical scaling and conservation laws give no a priori control of higher derivatives) for large initial data?

Main argument

The energy-critical NLW. Section 5.1 introduces the main object of the chapter: the energy-critical defocusing nonlinear wave equation $\partial{tt} u - \Delta u = -|u|^{4/(d-2)}u$ in $\mathbb{R}^{1+d}$, $d \geq 3$. This is the Strauss exponent; the nonlinearity is exactly $H^1$-critical (the energy $|\nabla u|{L^2}^2 + |u|_{L^{2^}}^{2^}$ is invariant under the natural scaling). For small initial data, global well-posedness and scattering follow from the perturbation lemma; the difficulty is large data. The chapter presents the strategy developed by Struwe (1988 in 3D, radial), Grillakis (1990, 3D), and Shatah–Struwe (1993–1994, general $d$) for the defocusing case.

Bubbles of energy concentration. Section 5.2 analyzes how energy can concentrate at isolated space-time points. If the solution $u$ were to blow up or fail to scatter, there would be a sequence of times $tn$ and points $xn$ at which the energy concentrates: $\liminf{n\to\infty}\int{|x-xn|\leq r}(|\nabla u(tn,x)|^2 + |u(tn,x)|^{2^*}) \geq \varepsilon0 > 0$ for each fixed $r > 0$ and some $\varepsilon_0$ independent of the scale. These are called energy bubbles or energy concentration scenarios. The argument shows that if the solution is not globally well-behaved, such bubbles must form.

Local Morawetz and non-concentration of mass. Section 5.3 proves the local Morawetz inequality for the energy-critical NLW: $\int0^T\int{\mathbb{R}^d}\frac{|u(t,x)|^{2d/(d-2)}}{|x|}\,dx\,dt \lesssim E[u0, u1]$. This is a spacetime weighted $L^2$-type estimate that controls how long the solution can remain concentrated near the spatial origin. The Morawetz inequality does not immediately give global control because it only bounds the contribution near the origin; the full argument shows concentration near any compact region is bounded.

Minimal-energy blowup solutions. Section 5.4 introduces the induction on energy method (Bourgain, 1999) and the related concentration-compactness method. The idea: if global well-posedness and scattering fail, there exists a minimal-energy blowup solution — a solution with the smallest possible energy that still fails to have the desired global properties. This minimal blowup solution is almost periodic (its spatial profile recurs in a rescaled sense), which severely constrains its geometry. Using the Morawetz inequality and the almost-periodicity, one derives a contradiction.

Global Morawetz and non-concentration of mass. Section 5.5 strengthens the Morawetz argument to a global Morawetz estimate in the defocusing case. Using the interaction Morawetz inequality (introduced by Colliander, Keel, Staffilani, Takaoka, Tao for NLS): $\int_0^T\int\int\frac{|u(t,x)|^2|u(t,y)|^2}{|x-y|}\,dx\,dy\,dt \lesssim E^2$, one proves that the solution cannot concentrate its $L^p$ norm in any compact region for long. Together with the small-data scattering theory and a Vitali covering argument, this gives global well-posedness and scattering for all finite-energy initial data (defocusing, energy-critical).

Key ideas

  • Energy criticality means the conservation law (energy) is invariant under scaling; there is no a priori size restriction on the Sobolev norm of the solution at late times.
  • Small-data global well-posedness and scattering follow from the perturbation lemma plus Strichartz; large-data arguments require a fundamentally different strategy.
  • The induction-on-energy method reduces the large-data problem to proving global scattering under the assumption that all solutions with strictly smaller energy scatter.
  • Minimal-energy blowup solutions are almost periodic; the Morawetz inequality then forces their $L^p$ norm to decay, contradicting non-trivial energy.
  • The interaction Morawetz inequality (a bilinear Morawetz estimate) provides the global spacetime $L^4$ control needed to close the defocusing large-data scattering argument for NLS.
  • The energy-critical NLW case $d=3$ (Struwe, Grillakis, Shatah–Struwe) is the historical prototype; Tao's presentation unifies this with the NLS case through the abstract induction-on-energy framework.

Key takeaway

Chapter 5 presents the state-of-the-art proof of large-data global well-posedness and scattering for energy-critical defocusing NLS and NLW, centered on the induction-on-energy / concentration-compactness method and the interaction Morawetz inequality — the deepest results in the book.

Chapter 6 — Wave Maps

Central question

How does one prove local well-posedness and, in subcritical regimes, global regularity for wave maps — solutions $u: \mathbb{R}^{1+d}\to M$ (where $M$ is a Riemannian manifold) satisfying the geometric wave equation $\Box u^\alpha + \Gamma^\alpha{\beta\gamma}(u)\partial^\mu u^\beta \partial\mu u^\gamma = 0$ — when the equation has gauge symmetry and cannot be handled by scalar techniques alone?

Main argument

Local theory. Section 6.1 sets up the wave maps equation as a system of semilinear wave equations with a quadratic nonlinearity in the first-order derivatives: $\Box u = Q(\partial u, \partial u)$ where $Q$ is a bilinear form determined by the Christoffel symbols of $M$. For the critical case $d=2$ (energy-critical wave maps), the small-data local-in-time theory follows from a Strichartz-space contraction argument. The chapter focuses on wave maps into compact Riemannian manifolds and into symmetric spaces.

Orthonormal frames and gauge transformations. Section 6.2 introduces the moving frame technique. Instead of analyzing the components $u^\alpha$ directly, one works with an orthonormal frame $(e1, \ldots, en)$ for the pullback bundle $u^TM$ along the map $u$. The connection 1-form $A = (A_\mu)$ encodes the geometry and satisfies a system of equations coupled to the frame. The *Coulomb gauge** condition $\partial^i A_i = 0$ diagonalizes the leading-order symbol, making the equation amenable to Fourier methods. In low dimensions the Coulomb gauge creates difficulties (the inverse of $-\Delta$ creates a non-integrable kernel), and Tao's caloric gauge (described briefly) resolves these.

Wave map decay estimates. Section 6.3 establishes decay for the connection 1-form and the frame in the subcritical regime $d \geq 3$. Using Strichartz estimates adapted to the wave equation (Chapter 2), one shows that the connection 1-form satisfies $L^p$ bounds that can be fed back into the equation for the frame. The key estimate is a product rule in $X^{s,b}$ spaces for the quadratic term $A\cdot\partial u$, showing it maps bounded sets to bounded sets in the appropriate Strichartz space.

Heat flow. Section 6.4 introduces the harmonic map heat flow $\partial_s u = \tau(u)$ (where $\tau$ is the tension field) as a technical tool for constructing the caloric gauge. Given a wave map $u(t,\cdot)$ at a fixed time $t$, one flows it toward a harmonic map via the heat flow and constructs an orthonormal frame along the heat flow that is parallel with respect to the heat flow parameter $s$. As $s\to\infty$, the frame converges to a fixed reference frame at the target harmonic map. Pulling back this "caloric" frame to $s=0$ gives the caloric gauge, in which the connection 1-form decays as $s\to\infty$ and the nonlinear term in the wave map equation becomes a nonlinear paraproduct (rather than an inverse derivative applied to a quadratic form), removing the low-dimensional obstruction.

Key ideas

  • Wave maps are the geometric generalization of scalar wave equations; the target geometry enters through the Christoffel symbols (equivalently, the second fundamental form for manifolds embedded in Euclidean space).
  • The moving frame reduces the geometric equation to a gauge-theoretic system: a connection (gauge field) and a section (frame) coupled together.
  • The Coulomb gauge simplifies the principal symbol but fails in low dimensions due to the non-integrable Riesz transform kernel.
  • The caloric gauge (constructed via the harmonic map heat flow) replaces the inverse Laplacian with a heat-kernel smoothing, making the nonlinear term more tractable at low regularity and in low dimensions.
  • Local well-posedness in energy-subcritical dimensions follows from Strichartz + Coulomb gauge; the energy-critical case ($d=2$) requires the caloric gauge or the Coulomb gauge on high-frequency components.
  • The results in this chapter are partial — large-data global regularity for energy-critical wave maps in $d=2$ was not yet fully established when the book was written (it was completed by Tao, Sterbenz–Tataru, and others in 2008–2011).

Key takeaway

Chapter 6 demonstrates that geometric PDE require additional structure beyond scalar Strichartz theory — specifically, moving frames and gauge choices — and that the caloric gauge provides the key technical tool for handling the critical 2D wave map problem.

Appendix A — Tools from Harmonic Analysis

Central question

What are the harmonic-analytic tools — Sobolev embeddings, Littlewood–Paley theory, Fourier restriction, and related results — that are used without proof in the main text, and how are they proved?

Main argument

Appendix A collects the background material in harmonic analysis that is invoked throughout the book but not derived in the main chapters. It is structured as a self-contained reference.

Sobolev spaces and embeddings. The Sobolev space $H^s = W^{s,2}$ is defined via the Fourier transform: $|f|_{H^s}^2 = \int \langle\xi\rangle^{2s}|\hat f(\xi)|^2\,d\xi$. The embedding $H^s \hookrightarrow L^{2^}$ where $2^ = 2d/(d-2s)$ (Sobolev embedding) is proved, as is the Gagliardo–Nirenberg interpolation inequality. These are used whenever one needs to control an $L^p$ norm from an $H^s$ norm in Chapters 3–6.

Littlewood–Paley theory. The Littlewood–Paley decomposition $f = \sumN PN f$ writes $f$ as a sum of frequency-localized pieces ($PN$ projects onto frequencies $|\xi| \sim N$). The Bernstein inequalities $|PN f|{L^q} \lesssim N^{d(1/p-1/q)}|PN f|{L^p}$ for $q \geq p$ exchange regularity for integrability on frequency-localized pieces. The Littlewood–Paley square function theorem gives $|f|{L^p} \sim |(\sumN |PN f|^2)^{1/2}|_{L^p}$ for $1 < p < \infty$. These tools are used in virtually every Strichartz and multilinear estimate in the book.

Fractional derivatives and the chain rule. The Kato–Ponce commutator estimate $|[D^s, f]g|{L^p} \lesssim |\nabla f|{L^{p1}}|D^{s-1}g|{L^{p2}} + |D^s f|{L^{q1}}|g|{L^{q2}}$ and the fractional chain rule $|D^s F(u)|{L^p} \lesssim |F'(u)|{L^{p1}}|D^s u|{L^{p2}}$ are proved. These allow one to commute fractional differential operators through compositions, which is necessary for proving energy estimates for $H^s$ solutions with $s \notin \mathbb{Z}$.

Hardy–Littlewood–Sobolev and potential theory. The Hardy–Littlewood–Sobolev inequality $||x|^{-\alpha}*f|{L^q} \lesssim |f|{L^p}$ (for appropriate $p, q, \alpha$) is stated and used to bound Riesz potentials. This enters the Morawetz estimates and gauge-fixing arguments.

Bilinear Strichartz estimates. Several bilinear refinements of Strichartz estimates are collected here, including the bilinear restriction estimates of Klainerman–Machedon and Bourgain that are used in the energy-critical and wave maps arguments.

Key takeaway

Appendix A makes the book self-contained for a reader with graduate real analysis: all Fourier-analytic tools — Sobolev, Littlewood–Paley, Bernstein, Kato–Ponce, Hardy–Littlewood–Sobolev — are proved in one place and can be referred back to as black boxes in the main text.

Appendix B — Construction of Ground States

Central question

What are the ground states (minimal-energy stationary solutions) of the NLS equation, how are they constructed, and why do they appear as the threshold objects in the focusing energy-critical scattering theory?

Main argument

Standing waves and the ground state NLS. The ground state for the focusing NLS is a positive radial function $Q \in H^1(\mathbb{R}^d)$ satisfying the nonlinear elliptic equation $-\Delta Q + Q = Q^{p}$ (for the $L^2$-critical or $H^1$-subcritical focusing NLS). It is characterized as the unique positive radial solution, up to translations and phase rotations. In the focusing $L^2$-critical case ($p = 1 + 4/d$), $Q$ is the Weinstein minimizer: the function minimizing the Gagliardo–Nirenberg functional $GN(u) = |\nabla u|{L^2}^d|u|{L^2}^{4-d(p-1)} / |u|_{L^{p+1}}^{p+1}$.

Existence via variational methods. The ground state is constructed via the method of Lagrange multipliers: minimize $|\nabla u|{L^2}^2$ subject to the constraint $|u|{L^{p+1}}^{p+1} = 1$ (or equivalently, minimize $|u|_{L^{p+1}}^{p+1}$ subject to fixed $H^1$ norm). Concentration-compactness of Lions guarantees that the minimizing sequence does not disperse to zero or split into multiple bumps — it concentrates on a single bump that, after rescaling, converges to $Q$.

Pohozaev identities. The Pohozaev identity is used to establish uniqueness and to derive necessary relations between the $L^2$, $H^1$, and $L^{p+1}$ norms of any solution to $-\Delta Q + Q = Q^p$: $|\nabla Q|{L^2}^2 = \frac{d(p-1)}{2(p+1)}|Q|{L^{p+1}}^{p+1}$. Combined with the Gagliardo–Nirenberg inequality, this gives the sharp constant in that inequality and shows that $Q$ is the unique optimizer.

Role in scattering theory. The ground state mass $M[Q] = |Q|{L^2}^2$ is the sharp threshold for the $L^2$-critical NLS: solutions with $M[u0] < M[Q]$ are global and scatter; solutions with $M[u_0] > M[Q]$ can blow up in finite time (the blowup solutions in the focusing case are rescalings of $Q$ itself). The appendix makes this threshold explicit and provides the self-contained existence and uniqueness result for $Q$ that is invoked in Chapter 5.

Key takeaway

Appendix B constructs the ground state $Q$ (the minimal-energy standing wave) via variational methods and establishes the Pohozaev identities, providing the sharp threshold object that separates global existence from blowup in the focusing energy-critical scattering theory.

The book's overall argument

  1. Chapter 1 (Ordinary Differential Equations) — establishes the complete toolkit (fixed-point iteration, Gronwall, bootstrap, Noether charges, Duhamel, monotonicity) in the simpler ODE setting, where the ideas are cleanest and generalization to PDE is direct.
  2. Chapter 2 (Constant Coefficient Linear Dispersive Equations) — introduces the Fourier-side description of dispersion, computes the fundamental solution, proves the Strichartz estimates that are the main quantitative output of dispersion, and sets up the $X^{s,b}$ spaces needed for KdV.
  3. Chapter 3 (Semilinear Dispersive Equations) — applies Chapters 1–2 to the nonlinear problem: scaling analysis identifies the critical regularity, the contraction argument gives local well-posedness, conservation laws give global results in the defocusing case, and the I-method extends global theory below energy regularity.
  4. Chapter 4 (The Korteweg–de Vries Equation) — treats an equation whose bilinear structure prevents direct Strichartz methods, requiring Bourgain spaces, the Miura transform, and gauge transforms; the complete integrability of KdV is exploited through its infinite conservation law hierarchy and Gromov-type symplectic non-squeezing.
  5. Chapter 5 (Energy-Critical Semilinear Dispersive Equations) — tackles the hardest case: large-data global theory at the scaling-invariant energy regularity, requiring the induction-on-energy / concentration-compactness method and the interaction Morawetz inequality to prove that minimal-energy blowup solutions cannot exist in the defocusing regime.
  6. Chapter 6 (Wave Maps) — moves to a geometric equation where the target is a manifold rather than $\mathbb{R}$; the scalar framework is replaced by moving frames and gauge theory, and the caloric gauge (constructed via the harmonic map heat flow) resolves the difficulties of the Coulomb gauge in low dimensions.
  7. Appendix A (Tools from Harmonic Analysis) — provides the self-contained harmonic analysis background (Sobolev, Littlewood–Paley, Bernstein, Kato–Ponce) that the main text invokes without proof.
  8. Appendix B (Construction of Ground States) — constructs the ground state $Q$ via variational methods, providing the sharp threshold that separates scattering from blowup in the focusing energy-critical NLS.

Common misunderstandings

Misunderstanding: The book covers all nonlinear dispersive equations.

The book explicitly concentrates on a representative sample of results for a selected set of equations. Many important equations (Dirac, Klein–Gordon, nonlinear Schrödinger with potentials, Schrödinger maps, MHD) are not treated. The selection is governed by the principle: choose equations that illustrate distinct technical phenomena (KdV for complete integrability and Bourgain spaces; wave maps for gauge theory) rather than maximal generality.

Misunderstanding: Dispersion always prevents blowup.

Dispersion is the tendency of waves to spread out, which generically prevents concentration. But for focusing equations, the nonlinearity can counteract dispersion. In the focusing NLS with $L^2$-supercritical nonlinearity, or in the focusing energy-critical NLS above the ground-state threshold, the nonlinearity wins and blowup occurs in finite time. The book carefully distinguishes focusing from defocusing throughout.

Misunderstanding: Local well-posedness is easy once one has Strichartz estimates.

Local well-posedness via Strichartz contraction requires the nonlinearity to be bounded in the appropriate Strichartz norm, which requires the nonlinearity to have the right power and regularity. For KdV, the standard contraction argument fails due to the derivative in the nonlinearity, requiring the $X^{s,b}$ framework. For wave maps, the gauge transformation is needed. The contraction argument is a template, not a push-button tool.

Misunderstanding: Scattering is a generic phenomenon for dispersive equations.

Scattering (solutions asymptote to free solutions at late times) holds for defocusing subcritical equations with small or $H^1$ data, and for defocusing energy-critical equations (proved in Chapter 5). It can fail for focusing equations (solitons do not scatter), for $L^2$-subcritical equations in low dimensions (modified scattering), and for solutions in dimensions where the linear decay rate is too slow.

Misunderstanding: The induction-on-energy method and concentration-compactness are the same thing.

They are related but distinct. Bourgain's induction-on-energy method (Chapter 5) proceeds by induction on the energy of the initial data, assuming the result for strictly smaller energies and deriving it for the current energy. Concentration-compactness (Lions, Kenig–Merle) extracts a minimal-energy counterexample directly from a failing sequence. Both produce an almost-periodic minimal blowup solution, but the logical structure differs; Tao's chapter 5 uses the induction-on-energy framework.

Central paradox / key insight

The central paradox of the subject is that nonlinear dispersive equations can have solutions that exist globally in time despite strong nonlinear interactions, not because the nonlinearity is weak, but because dispersion and nonlinearity collaborate in a precise way.

For defocusing equations: the nonlinearity, while capable of concentrating energy, is ultimately controlled by the dispersive spreading. The Morawetz estimates quantify this: no matter how large the initial energy, the solution cannot concentrate in any bounded spacetime region for too long. The energy is forced to spread out by dispersion, and once it spreads, the nonlinearity becomes small. This feedback loop — dispersion forces spreading, spreading weakens the nonlinearity, which allows further dispersive spreading — is the underlying mechanism of the large-data global results in Chapter 5.

For focusing equations near the ground state threshold: the ground state $Q$ is the exact equilibrium where nonlinearity and dispersion balance. Below the threshold $M[u_0] < M[Q]$, dispersion wins and solutions scatter. Above the threshold, solutions can form solitons (traveling ground states) or blow up, depending on the geometry of the solution. This threshold is as sharp as possible: the ground state itself is a non-scattering, non-blowing-up stationary solution sitting precisely on the dividing line.

The deepest results in the book show that, for the energy-critical defocusing equations, no concentration scenario can persist: the interaction Morawetz inequality forces the nonlinear $L^4$ spacetime norm to be finite, which is precisely what scattering requires.

Important concepts

Dispersion relation

The relationship $\tau = h(\xi)$ between temporal frequency $\tau$ and spatial frequency $\xi$ for a linear constant-coefficient PDE. The equation is dispersive when $\nabla^2\xi h \neq 0$, meaning the group velocity $\nabla\xi h$ is non-constant and different frequency components travel at different speeds.

Strichartz estimates

Space-time mixed-norm bounds for the free evolution: $|e^{it\Delta}u0|{L^pt L^qx(\mathbb{R}^{1+d})} \lesssim |u0|{L^2(\mathbb{R}^d)}$ where the exponents $(p,q)$ satisfy the admissibility condition $\frac{2}{p} + \frac{d}{q} = \frac{d}{2}$, $p \geq 2$. They encode the dispersive decay of the linear propagator in an $L^2$-friendly form, and are the main tool for proving local well-posedness via contraction.

Critical Sobolev exponent $s_c$

For the NLS $i\partialt u + \Delta u = \lambda|u|^{p-1}u$ in $\mathbb{R}^d$, the Sobolev index $sc = d/2 - 2/(p-1)$ that is invariant under the natural scaling. Sub-critical regularity ($s > sc$) allows perturbative local well-posedness; critical regularity ($s = sc$) gives scale-invariant well-posedness; super-critical regularity ($s < s_c$) defeats perturbative methods.

$X^{s,b}$ spaces (Bourgain spaces)

Function spaces on $\mathbb{R}^{1+d}$ defined by $|u|_{X^{s,b}}^2 = \int \langle\xi\rangle^{2s}\langle\tau - h(\xi)\rangle^{2b}|\hat u(\tau,\xi)|^2\,d\tau\,d\xi$, where $h(\xi)$ is the dispersion relation. They measure how well a function concentrates near the characteristic surface of the associated linear equation, and are the natural spaces for bilinear estimates for KdV.

Duhamel formula

The integral representation $u(t) = e^{itP(D)}u0 + \int0^t e^{i(t-s)P(D)}F(u(s))\,ds$ that converts the Cauchy problem for a linear PDE plus forcing into a Volterra integral equation. It is the starting point for every fixed-point (Picard iteration) proof of local well-posedness.

Bootstrap principle

If a property $P(t)$ on $[0, T^]$ is (a) open, (b) closed relative to the assumption that $P$ holds on a slightly weaker property, and (c) holds at $t = 0$, then $P(t)$ holds for all $t \in [0, T^]$. Used to close a priori estimates when the bound to be proved appears on both sides of an inequality.

Morawetz inequality

A spacetime integral estimate of the form $\int0^T\int{\mathbb{R}^d} a(x)|u(t,x)|^p\,dx\,dt \lesssim E[u_0]$ that provides one-sided (non-conservative) control on the solution. The weight $a(x)$ (typically $1/|x|$) controls the contribution near a fixed spatial location. For energy-critical NLS, the interaction Morawetz estimate $\int\int\int \frac{|u(t,x)|^2|u(t,y)|^2}{|x-y|}\,dx\,dy\,dt \lesssim E^2$ is the key to large-data scattering.

Induction on energy / concentration-compactness

A method for proving large-data global results by assuming the desired result (scattering) holds for all solutions with energy strictly below a threshold $E$, and deriving a contradiction at energy level $E$ by showing a hypothetical non-scattering minimal-energy solution must be almost periodic, which the Morawetz inequality forbids.

Ground state $Q$

The unique (up to symmetries) positive radial solution $Q \in H^1(\mathbb{R}^d)$ to $-\Delta Q + Q = Q^p$ (for the focusing NLS). It minimizes the Gagliardo–Nirenberg functional and its $L^2$ norm $M[Q]$ is the sharp threshold: solutions below $M[Q]$ scatter, solutions above $M[Q]$ can blow up.

Miura transform

The substitution $u = vx + v^2$ mapping solutions of the focusing modified KdV $\partialt v + \partial{xxx}v - 6v^2\partialx v = 0$ to solutions of KdV $\partialt u + \partial{xxx}u + 6u\partial_x u = 0$. It allows results proved for mKdV (which has better conservation laws) to be transferred to KdV.

Wave maps equation

The equation $\Box u^\alpha + \Gamma^\alpha{\beta\gamma}(u)\partial^\mu u^\beta \partial\mu u^\gamma = 0$ for a map $u: \mathbb{R}^{1+d}\to M$ from Minkowski space to a Riemannian manifold $(M,g)$. It is the Lorentzian analogue of the harmonic map equation and includes NLS/NLW as special cases when $M = \mathbb{R}$ or $S^1$.

Caloric gauge

A choice of orthonormal frame for the pullback bundle $u^*TM$ along a wave map, constructed by flowing $u(t,\cdot)$ along the harmonic map heat flow to a reference harmonic map and pulling back a parallel frame. In this gauge, the connection 1-form decays as the heat parameter $s\to\infty$, and the nonlinear term in the wave map system becomes a paraproduct rather than an inverse derivative applied to a quadratic, removing the low-dimensional obstruction of the Coulomb gauge.

I-method (almost conservation laws)

A method for proving global well-posedness for NLS below the energy regularity by replacing the exact energy with a modified energy $E[Iu]$, where $I$ is a frequency cutoff smoothing operator. The modified energy is not exactly conserved but its time derivative is $O(N^{-\alpha})$ for large frequency $N$; iterating local results over $O(N^\alpha)$ time intervals gives global solutions.

Primary book and edition information

Author's page and errata

Background and overview

Foundational results the book builds on

  • Keel, M. and Tao, T. "Endpoint Strichartz estimates." American Journal of Mathematics 120 (1998), 955–980. — the endpoint Strichartz estimates (Chapter 2, Section 2.3).
  • Colliander, J., Keel, M., Staffilani, G., Takaoka, H., and Tao, T. "Sharp global well-posedness for KdV and modified KdV on $\mathbb{R}$ and $\mathbb{T}$." Journal of the American Mathematical Society 16 (2003), 705–749.
  • Bourgain, J. "Global wellposedness of defocusing critical nonlinear Schrödinger equation in the radial case." Journal of the American Mathematical Society 12 (1999), 145–171. — the induction-on-energy method (Chapter 5).
  • Kenig, C. E. and Merle, F. "Global well-posedness, scattering and blow-up for the energy-critical, focusing, non-linear Schrödinger equation in the radial case." Inventiones Mathematicae 166 (2006), 645–675. — concentration-compactness and ground-state threshold (Chapter 5, Appendix B).

Additional study resources

These secondary records are supplements to the primary text and should not be used as substitutes for the book itself.

Send feedback

Optional. We'll only use this if you want a reply.