Intro Previous work Main results Geometric formulation Estimates Energy hierarchy An Overview of Recent Progress on Shock Formation in Three Spatial Dimensions Jared Speck with G. Holzegel, S. Klainerman, & W. Wong Massachusetts Institute of Technology Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Maximal developments Theorem (Classic, Sbierski, Wong) Consider the initial value problem m φ = N (φ, ∂φ, ∂ 2 φ) := Nonlinearity, φ|Σ0 = φ̊, ∂t φ|Σ0 = φ̊0 , on R1+3 , where m = diag(−1, 1, 1, 1), m := −∂t2 + ∆, Σ0 is a spacelike hypersurface, and (φ̊, φ̊0 ) are smooth. Then ∃! maximal development. What is the nature of the maximal development? Global existence? Singularities? Cauchy horizon? Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Maximal developments Theorem (Classic, Sbierski, Wong) Consider the initial value problem m φ = N (φ, ∂φ, ∂ 2 φ) := Nonlinearity, φ|Σ0 = φ̊, ∂t φ|Σ0 = φ̊0 , on R1+3 , where m = diag(−1, 1, 1, 1), m := −∂t2 + ∆, Σ0 is a spacelike hypersurface, and (φ̊, φ̊0 ) are smooth. Then ∃! maximal development. What is the nature of the maximal development? Global existence? Singularities? Cauchy horizon? Intro Previous work Main results Geometric formulation Estimates Energy hierarchy A maximal development with a singular boundary boundary of maximal development 6 t Σ0 = spacelike Intro Previous work Main results Geometric formulation Estimates Energy hierarchy A model problem in 1D Burgers’ equation for Ψ(t, x): ∂t Ψ + Ψ∂x Ψ = 0 ˚ = data size; data := Ψ(0, x) It is famously known that shock singularities form at TShock = O(˚ −1 ); Ψ remains bounded but |∂x Ψ| → ∞ A much less appreciated fact is that: The singularity is renormalizable. That is, the solution is globally smooth relative to good coordinates. Intro Previous work Main results Geometric formulation Estimates Energy hierarchy A model problem in 1D Burgers’ equation for Ψ(t, x): ∂t Ψ + Ψ∂x Ψ = 0 ˚ = data size; data := Ψ(0, x) It is famously known that shock singularities form at TShock = O(˚ −1 ); Ψ remains bounded but |∂x Ψ| → ∞ A much less appreciated fact is that: The singularity is renormalizable. That is, the solution is globally smooth relative to good coordinates. Intro Previous work Main results Geometric formulation Estimates Energy hierarchy A model problem in 1D Burgers’ equation for Ψ(t, x): ∂t Ψ + Ψ∂x Ψ = 0 ˚ = data size; data := Ψ(0, x) It is famously known that shock singularities form at TShock = O(˚ −1 ); Ψ remains bounded but |∂x Ψ| → ∞ A much less appreciated fact is that: The singularity is renormalizable. That is, the solution is globally smooth relative to good coordinates. Intro Previous work Main results Geometric formulation Estimates Energy hierarchy “Hiding” the singularity Characteristics: x = x(t; u) solves d x(t; u) = Ψ(t, x(t; u)), dt x(0; u) = u In characteristic coordinates (t, u), Burgers’ equation is ∂ Ψ(t, u) = 0 ∂t Thus, Ψ(t, u) = F (u) Intro Previous work Main results Geometric formulation Estimates Energy hierarchy “Hiding” the singularity Characteristics: x = x(t; u) solves d x(t; u) = Ψ(t, x(t; u)), dt x(0; u) = u In characteristic coordinates (t, u), Burgers’ equation is ∂ Ψ(t, u) = 0 ∂t Thus, Ψ(t, u) = F (u) Intro Previous work Main results Geometric formulation Estimates Shocks in Burgers’ equation solutions TShock u ≡ const ↓ µ := 1/∂x u = 0 ↓ = O(˚ −1 ) ˚ = data size t x Energy hierarchy Intro Previous work Main results Geometric formulation Estimates Energy hierarchy The inverse foliation density µ Set µ := 1 ∂x u Evolution equation for µ: ∂ ∂ µ(t, u) = Ψ(t, u) = F 0 (u) ∂t ∂u CHOV relation =⇒ ∂x Ψ blows up when µ → 0 : ∂x Ψ = 1 ∂ 1 Ψ = F0 µ ∂u µ Intro Previous work Main results Geometric formulation Estimates Energy hierarchy The inverse foliation density µ Set µ := 1 ∂x u Evolution equation for µ: ∂ ∂ µ(t, u) = Ψ(t, u) = F 0 (u) ∂t ∂u CHOV relation =⇒ ∂x Ψ blows up when µ → 0 : ∂x Ψ = 1 ∂ 1 Ψ = F0 µ ∂u µ Intro Previous work Main results Geometric formulation Estimates Energy hierarchy The inverse foliation density µ Set µ := 1 ∂x u Evolution equation for µ: ∂ ∂ µ(t, u) = Ψ(t, u) = F 0 (u) ∂t ∂u CHOV relation =⇒ ∂x Ψ blows up when µ → 0 : ∂x Ψ = 1 ∂ 1 Ψ = F0 µ ∂u µ Intro Previous work Main results Geometric formulation Estimates The 1D theory There are a huge number of related results 1D : Challis (1848) ··· Zabusky (1962) Lax (1964) Keller and Ting (1966) John (1974) Klainerman and Majda (1980) Majda (1984) ··· Energy hierarchy Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Non-constructive breakdown in 3D Theorem (John 1980) In R1+3 , consider nontrivial initially smooth solutions to ( −(∂t φ)2 , m φ = −(∂t φ)∂t2 φ φ|t=0 = φ̊, , ∂t φ|t=0 = φ̊0 , where m = diag(−1, 1, 1, 1) and m := −∂t2 + ∆. Then φ breaks down for some unknown reason. A different kind of non-constructive blow-up proof for the Euler equations was provided by Sideris and later by Guo+Tahvildar-Zadeh. Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Non-constructive breakdown in 3D Theorem (John 1980) In R1+3 , consider nontrivial initially smooth solutions to ( −(∂t φ)2 , m φ = −(∂t φ)∂t2 φ φ|t=0 = φ̊, , ∂t φ|t=0 = φ̊0 , where m = diag(−1, 1, 1, 1) and m := −∂t2 + ∆. Then φ breaks down for some unknown reason. A different kind of non-constructive blow-up proof for the Euler equations was provided by Sideris and later by Guo+Tahvildar-Zadeh. Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Null condition in 3D In the previous examples, S. Klainerman’s classic null condition fails; there are quadratic terms proportional to (∇LFlat φ)2 , ∇LFlat φ · ∇2LFlat φ t t− LF lat = ∂t − ∂r r= LF lat = ∂t + ∂r co ns t angular directions Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Relevance of null condition Theorem (Christodoulou, Klainerman 1984) For data of small size ˚ , solutions to m φ = Q1 (∂φ, ∂φ) + Q2 (∂ 2 φ, ∂φ) exist globally when the Qi verify the classic null condition. Example: Q1 (∂φ, ∂φ) = (∂t φ)2 − (∂r φ)2 = ∇LFlat φ · ∇LFlat φ Key ingredient: linear solutions m φ = 0 disperse: |∇LFlat φ| . ˚ 1 , 1+t |∇LFlat φ|, |∇ / φ| . ˚ 1 (1 + t)2 Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Relevance of null condition Theorem (Christodoulou, Klainerman 1984) For data of small size ˚ , solutions to m φ = Q1 (∂φ, ∂φ) + Q2 (∂ 2 φ, ∂φ) exist globally when the Qi verify the classic null condition. Example: Q1 (∂φ, ∂φ) = (∂t φ)2 − (∂r φ)2 = ∇LFlat φ · ∇LFlat φ Key ingredient: linear solutions m φ = 0 disperse: |∇LFlat φ| . ˚ 1 , 1+t |∇LFlat φ|, |∇ / φ| . ˚ 1 (1 + t)2 Intro Previous work Main results Geometric formulation Estimates Energy hierarchy John’s proof of spherically symmetric blow-up Theorem (John 1985) In 3D, consider nontrivial spherically symmetric data of small size ˚ for −∂t2 φ + (1 + ∂t φ)∆φ = 0 Then ∇L ∂r φ blows up at TLifespan = O(exp(c˚ −1 )). φ and its first rectangular derivatives remain bounded. Blow-up follows from a Riccati inequality for w ∼ r ∇L ∂r φ along characteristics: 1 d w ≥c w 2 + Error dt 1+t Intro Previous work Main results Geometric formulation Estimates Energy hierarchy John’s proof of spherically symmetric blow-up Theorem (John 1985) In 3D, consider nontrivial spherically symmetric data of small size ˚ for −∂t2 φ + (1 + ∂t φ)∆φ = 0 Then ∇L ∂r φ blows up at TLifespan = O(exp(c˚ −1 )). φ and its first rectangular derivatives remain bounded. Blow-up follows from a Riccati inequality for w ∼ r ∇L ∂r φ along characteristics: 1 d w ≥c w 2 + Error dt 1+t Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Main difficulties away from symmetry Away from symmetry, the term Error in d 1 w ≥c w 2 + Error dt 1+t depends on the angular Laplacian ∆ / φ. Angular derivatives can only be controlled with energy estimates. Suitable energy estimates showing that ∇ / φ = Error are very hard to derive near the blow-up points. The Riccati inequality approach is not precise enough to work. Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Main difficulties away from symmetry Away from symmetry, the term Error in d 1 w ≥c w 2 + Error dt 1+t depends on the angular Laplacian ∆ / φ. Angular derivatives can only be controlled with energy estimates. Suitable energy estimates showing that ∇ / φ = Error are very hard to derive near the blow-up points. The Riccati inequality approach is not precise enough to work. Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Alinhac’s shock formation results Theorem (Alinhac 1990s-early 2000s) In 3D, consider nontrivial data of small size ˚ for (g −1 )αβ (∂φ)∂α ∂β φ = 0, gαβ (∂φ) = mαβ + (Einstein summation) Pert (∂φ), gαβ Pert (∂φ = 0) = 0 gαβ Assume that the null condition fails. Then under a non-degeneracy assumption on the data: ∃ exactly one blow-up point at TShock = O(exp(c˚ −1 )). Singularity can be partially renormalized. Blow-up is tied to intersecting characteristics. Extensions by Huicheng, Bingbing, Witt Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Alinhac’s shock formation results Theorem (Alinhac 1990s-early 2000s) In 3D, consider nontrivial data of small size ˚ for (g −1 )αβ (∂φ)∂α ∂β φ = 0, gαβ (∂φ) = mαβ + (Einstein summation) Pert (∂φ), gαβ Pert (∂φ = 0) = 0 gαβ Assume that the null condition fails. Then under a non-degeneracy assumption on the data: ∃ exactly one blow-up point at TShock = O(exp(c˚ −1 )). Singularity can be partially renormalized. Blow-up is tied to intersecting characteristics. Extensions by Huicheng, Bingbing, Witt Intro Previous work Main results Geometric formulation Estimates Energy hierarchy The region studied by Alinhac St,0 Σut St,u 6 t S0,u Σu0 S0,0 nontrivial data Cut C0t trivial data φ≡0 Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Christodoulou’s shock formation results in 3D In his famous book of 2007, Christodoulou (and with Miao in 2013) proved a related small-data blow-up result. His proof yielded valuable new information that is inaccessible via Alinhac’s approach. He studied perturbations of the solution φ ≡ t to all Euler-Lagrange PDEs from irrotational fluid mechanics. Example equations are the following with s > 1: ∂α ∂σs ∂(∂α φ) = 0, σ := −(m−1 )κλ ∂κ φ∂λ φ Singularities form ⇐⇒ characteristics intersect. To generate shocks, he made assumptions on Z Z r {(∂t φ − 1) − c∂r φ} dσ + 2(∂t φ − 1) − c∂r φ d 3 x S0,u Σu0 Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Christodoulou’s shock formation results in 3D In his famous book of 2007, Christodoulou (and with Miao in 2013) proved a related small-data blow-up result. His proof yielded valuable new information that is inaccessible via Alinhac’s approach. He studied perturbations of the solution φ ≡ t to all Euler-Lagrange PDEs from irrotational fluid mechanics. Example equations are the following with s > 1: ∂α ∂σs ∂(∂α φ) = 0, σ := −(m−1 )κλ ∂κ φ∂λ φ Singularities form ⇐⇒ characteristics intersect. To generate shocks, he made assumptions on Z Z r {(∂t φ − 1) − c∂r φ} dσ + 2(∂t φ − 1) − c∂r φ d 3 x S0,u Σu0 Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Christodoulou’s shock formation results in 3D In his famous book of 2007, Christodoulou (and with Miao in 2013) proved a related small-data blow-up result. His proof yielded valuable new information that is inaccessible via Alinhac’s approach. He studied perturbations of the solution φ ≡ t to all Euler-Lagrange PDEs from irrotational fluid mechanics. Example equations are the following with s > 1: ∂α ∂σs ∂(∂α φ) = 0, σ := −(m−1 )κλ ∂κ φ∂λ φ Singularities form ⇐⇒ characteristics intersect. To generate shocks, he made assumptions on Z Z r {(∂t φ − 1) − c∂r φ} dσ + 2(∂t φ − 1) − c∂r φ d 3 x S0,u Σu0 Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Christodoulou’s shock formation results in 3D In his famous book of 2007, Christodoulou (and with Miao in 2013) proved a related small-data blow-up result. His proof yielded valuable new information that is inaccessible via Alinhac’s approach. He studied perturbations of the solution φ ≡ t to all Euler-Lagrange PDEs from irrotational fluid mechanics. Example equations are the following with s > 1: ∂α ∂σs ∂(∂α φ) = 0, σ := −(m−1 )κλ ∂κ φ∂λ φ Singularities form ⇐⇒ characteristics intersect. To generate shocks, he made assumptions on Z Z r {(∂t φ − 1) − c∂r φ} dσ + 2(∂t φ − 1) − c∂r φ d 3 x S0,u Σu0 Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Christodoulou’s shock formation results in 3D In his famous book of 2007, Christodoulou (and with Miao in 2013) proved a related small-data blow-up result. His proof yielded valuable new information that is inaccessible via Alinhac’s approach. He studied perturbations of the solution φ ≡ t to all Euler-Lagrange PDEs from irrotational fluid mechanics. Example equations are the following with s > 1: ∂α ∂σs ∂(∂α φ) = 0, σ := −(m−1 )κλ ∂κ φ∂λ φ Singularities form ⇐⇒ characteristics intersect. To generate shocks, he made assumptions on Z Z r {(∂t φ − 1) − c∂r φ} dσ + 2(∂t φ − 1) − c∂r φ d 3 x S0,u Σu0 Intro Previous work Main results Geometric formulation Estimates Energy hierarchy New information in Christodoulou’s proof singular H C regular H ∂− H Figure: The maximal development Intro Previous work Main results Geometric formulation Estimates Energy hierarchy New shock formation results - Part I Theorem (G.H.,S.K.,J.S.,W.W.) Consider the initial value problem g(Ψ) Ψ := (g −1 )αβ (Ψ)∂α ∂β Ψ − Γα (Ψ, ∂Ψ)∂α Ψ = 0, Ψ|t=0 = Ψ̊, ∂t Ψ|t=0 = Ψ̊0 where (Ψ̊, Ψ̊0 ) are compactly supported and of small size ˚ . Assume that Pert Pert gαβ (Ψ) = mαβ + gαβ (Ψ), gαβ (0) = 0 Classic null condition fails We provide a complete description of the dynamics. Singularity ⇐⇒ characteristics intersect. Rescaling (Ψ̊, Ψ̊0 ) → λ(Ψ̊, Ψ̊0 ) for small λ always leads to a shock at TShock ∼ exp(c˚ −1 ). Intro Previous work Main results Geometric formulation Estimates Energy hierarchy New shock formation results - Part I Theorem (G.H.,S.K.,J.S.,W.W.) Consider the initial value problem g(Ψ) Ψ := (g −1 )αβ (Ψ)∂α ∂β Ψ − Γα (Ψ, ∂Ψ)∂α Ψ = 0, Ψ|t=0 = Ψ̊, ∂t Ψ|t=0 = Ψ̊0 where (Ψ̊, Ψ̊0 ) are compactly supported and of small size ˚ . Assume that Pert Pert gαβ (Ψ) = mαβ + gαβ (Ψ), gαβ (0) = 0 Classic null condition fails We provide a complete description of the dynamics. Singularity ⇐⇒ characteristics intersect. Rescaling (Ψ̊, Ψ̊0 ) → λ(Ψ̊, Ψ̊0 ) for small λ always leads to a shock at TShock ∼ exp(c˚ −1 ). Intro Previous work Main results Geometric formulation Estimates Energy hierarchy New shock formation results - Part II Corollary (G.H.,S.K.,J.S.,W.W.) The same result holds for Alinhac’s equations (g −1 )αβ (∂φ)∂α ∂β φ = 0, Pert gαβ = mαβ + gαβ (∂φ), Pert gαβ (∂φ = 0) = 0 Remark Miao-Yu recently obtained 2 a related large-data blow-up 2 result for − 1 + (∂t φ) ∂t φ + ∆φ = 0 Intro Previous work Main results Geometric formulation Estimates Energy hierarchy New shock formation results - Part II Corollary (G.H.,S.K.,J.S.,W.W.) The same result holds for Alinhac’s equations (g −1 )αβ (∂φ)∂α ∂β φ = 0, Pert gαβ = mαβ + gαβ (∂φ), Pert gαβ (∂φ = 0) = 0 Remark Miao-Yu recently obtained 2 a related large-data blow-up 2 result for − 1 + (∂t φ) ∂t φ + ∆φ = 0 Intro Previous work Main results Geometric formulation Estimates Energy hierarchy The three main features of the proof The singularity is renormalizable, just like in Burgers’ equation! Unlike in 1D, basic existence theory requires estimates for higher derivatives. Moreover, the first statement is false for the high derivatives: They are not renormalizable! This leads to severe complications! We must construct a sharp geometric framework to see the renormalizability and to overcome the difficulties. Intro Previous work Main results Geometric formulation Estimates Energy hierarchy The three main features of the proof The singularity is renormalizable, just like in Burgers’ equation! Unlike in 1D, basic existence theory requires estimates for higher derivatives. Moreover, the first statement is false for the high derivatives: They are not renormalizable! This leads to severe complications! We must construct a sharp geometric framework to see the renormalizability and to overcome the difficulties. Intro Previous work Main results Geometric formulation Estimates Energy hierarchy The three main features of the proof The singularity is renormalizable, just like in Burgers’ equation! Unlike in 1D, basic existence theory requires estimates for higher derivatives. Moreover, the first statement is false for the high derivatives: They are not renormalizable! This leads to severe complications! We must construct a sharp geometric framework to see the renormalizability and to overcome the difficulties. Intro Previous work Main results Geometric formulation Estimates Energy hierarchy The three main features of the proof The singularity is renormalizable, just like in Burgers’ equation! Unlike in 1D, basic existence theory requires estimates for higher derivatives. Moreover, the first statement is false for the high derivatives: They are not renormalizable! This leads to severe complications! We must construct a sharp geometric framework to see the renormalizability and to overcome the difficulties. Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Eikonal function Most important proof ingredient is the eikonal function: Du · Du := (g −1 )αβ (Ψ)∂α u∂β u = 0, ∂t u > 0, u|t=0 = 1 − r , Truncated level sets are null cone pieces Cut Play a critical role in many delicate local and global results for wave equations, especially Einstein’s equations (starting with the Christodoulou-Klainerman proof of the stability of Minkowski spacetime) Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Fundamental geometric constructions St,u 6 t C0t Cut R̆ L X ,X µ≈1 1 2 R̆ L µ is small X1 , X2 µ−1 := −Dt · Du, shock ⇐⇒ µ → 0 Lν := −µDν u, L ∼ ∂t + ∂r R̆u = 1, g(R̆, R̆) = µ2 , shock ⇐⇒ R̆ → 0 X1 , X2 are local vectorfields on St,u = Σt ∩ Cut Intro Previous work Main results Geometric formulation Estimates The system of equations g(Ψ) Ψ = 0, 1 Lµ = GLL R̆Ψ + Error(LΨ, ∇ / Ψ, · · · ), 2 LLi = Error(LΨ, ∇ / Ψ, · · · ) Last two equations are ∼ Du · Du = 0 ∇ / := (true) angular derivatives L = ∂t∂ |u,ϑ ∂ | + angular error ∂u t,ϑ d GLL := dΨ gαβ (Ψ)Lα Lβ R̆ = Classic null condition fails =⇒ GLL ∼ f (ϑ) 6= 0 Energy hierarchy Intro Previous work Main results Geometric formulation Estimates The system of equations g(Ψ) Ψ = 0, 1 Lµ = GLL R̆Ψ + Error(LΨ, ∇ / Ψ, · · · ), 2 LLi = Error(LΨ, ∇ / Ψ, · · · ) Last two equations are ∼ Du · Du = 0 ∇ / := (true) angular derivatives L = ∂t∂ |u,ϑ ∂ | + angular error ∂u t,ϑ d GLL := dΨ gαβ (Ψ)Lα Lβ R̆ = Classic null condition fails =⇒ GLL ∼ f (ϑ) 6= 0 Energy hierarchy Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Behavior of lower-order derivatives •˚ = data size Bootstrap “linear” behavior relative to the rescaled frame Relative to the rescaled frame, we should see the dispersive behavior of linear waves: 1 , (1 + t)2 1 |Ψ|, |R̆Ψ| . ˚ 1+t |LΨ|, |∇ / Ψ| . ˚ 1 Can prove: R̆Ψ ∼ fData (u, ϑ) 1+t , fData = O(˚ ) 6≡ 0 Captures the renormalizability of the singularity Is essential for controlling error terms Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Behavior of lower-order derivatives •˚ = data size Bootstrap “linear” behavior relative to the rescaled frame Relative to the rescaled frame, we should see the dispersive behavior of linear waves: 1 , (1 + t)2 1 |Ψ|, |R̆Ψ| . ˚ 1+t |LΨ|, |∇ / Ψ| . ˚ 1 Can prove: R̆Ψ ∼ fData (u, ϑ) 1+t , fData = O(˚ ) 6≡ 0 Captures the renormalizability of the singularity Is essential for controlling error terms Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Behavior of lower-order derivatives •˚ = data size Bootstrap “linear” behavior relative to the rescaled frame Relative to the rescaled frame, we should see the dispersive behavior of linear waves: 1 , (1 + t)2 1 |Ψ|, |R̆Ψ| . ˚ 1+t |LΨ|, |∇ / Ψ| . ˚ 1 Can prove: R̆Ψ ∼ fData (u, ϑ) 1+t , fData = O(˚ ) 6≡ 0 Captures the renormalizability of the singularity Is essential for controlling error terms Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Strong null condition Set %(t, u) := 1 − u + t ≈ 1 + t. Then µg Ψ = 0 is: n o L µL(%Ψ) + 2R̆(%Ψ) = %µ∆ / Ψ + Error • Key structure: Error consists of quadratic and higher terms, where (R̆Ψ)2 , (R̆Ψ)3 , etc. are not present. • This is a nonlinear analog of the classic null condition. “Strong null condition” Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Strong null condition Set %(t, u) := 1 − u + t ≈ 1 + t. Then µg Ψ = 0 is: n o L µL(%Ψ) + 2R̆(%Ψ) = %µ∆ / Ψ + Error • Key structure: Error consists of quadratic and higher terms, where (R̆Ψ)2 , (R̆Ψ)3 , etc. are not present. • This is a nonlinear analog of the classic null condition. “Strong null condition” Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Heuristics behind µ → 0 Lµ(t, u, ϑ) ∼ R̆Ψ(t, u, ϑ) From R̆Ψ(t, u, ϑ) ∼ 1 f (u, ϑ), 1+t Data Lµ(t, u, ϑ) ∼ Integrating along L = ∂ ∂t we obtain: 1 fData (u, ϑ) 1+t and using µ|t=0 ∼ 1, we conclude: µ(t, u, ϑ) ∼ 1 + ln(1 + t)fData (u, ϑ) Hence, µ → 0 at time TShock ∼ exp | min u,ϑ 1 fData (u, ϑ) | Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Heuristics behind µ → 0 Lµ(t, u, ϑ) ∼ R̆Ψ(t, u, ϑ) From R̆Ψ(t, u, ϑ) ∼ 1 f (u, ϑ), 1+t Data Lµ(t, u, ϑ) ∼ Integrating along L = ∂ ∂t we obtain: 1 fData (u, ϑ) 1+t and using µ|t=0 ∼ 1, we conclude: µ(t, u, ϑ) ∼ 1 + ln(1 + t)fData (u, ϑ) Hence, µ → 0 at time TShock ∼ exp | min u,ϑ 1 fData (u, ϑ) | Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Heuristics behind µ → 0 Lµ(t, u, ϑ) ∼ R̆Ψ(t, u, ϑ) From R̆Ψ(t, u, ϑ) ∼ 1 f (u, ϑ), 1+t Data Lµ(t, u, ϑ) ∼ Integrating along L = ∂ ∂t we obtain: 1 fData (u, ϑ) 1+t and using µ|t=0 ∼ 1, we conclude: µ(t, u, ϑ) ∼ 1 + ln(1 + t)fData (u, ϑ) Hence, µ → 0 at time TShock ∼ exp | min u,ϑ 1 fData (u, ϑ) | Intro Previous work Main results Geometric formulation Estimates Justifying the heuristics We need to show that Error terms are small Solution survives until shock Big difficulty: only L2 regularity is propagated at top order. Two main ingredients: A priori L2 estimates for Ψ, µ, Li | {z } hard at top order! k A priori C estimates for Ψ, µ, Li Energy hierarchy Intro Previous work Main results Geometric formulation Estimates Justifying the heuristics We need to show that Error terms are small Solution survives until shock Big difficulty: only L2 regularity is propagated at top order. Two main ingredients: A priori L2 estimates for Ψ, µ, Li | {z } hard at top order! k A priori C estimates for Ψ, µ, Li Energy hierarchy Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Main a priori L2 estimate hierarchy (hard!) Let ˚ := data size, µ? (t, u) := min{1, minΣut µ}, EM (t, u) := 2 L (Σut )−based energy of M derivatives of Ψ. Then, ignoring ln t factors at top-order, we have: E24 (t, u) . ˚ 2 µ−17.5 (t, u) ? E23 (t, u) . ˚ 2 µ−15.5 (t, u) ? ··· E16 (t, u) . ˚ 2 µ−1.5 (t, u) ? 2 E15 (t, u) . ˚ ··· E0 (t, u) . ˚ 2 We recover C k bounds for Ψ via Sobolev We recover C k bounds for µ, Li from the eikonal equation Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Main a priori L2 estimate hierarchy (hard!) Let ˚ := data size, µ? (t, u) := min{1, minΣut µ}, EM (t, u) := 2 L (Σut )−based energy of M derivatives of Ψ. Then, ignoring ln t factors at top-order, we have: E24 (t, u) . ˚ 2 µ−17.5 (t, u) ? E23 (t, u) . ˚ 2 µ−15.5 (t, u) ? ··· E16 (t, u) . ˚ 2 µ−1.5 (t, u) ? 2 E15 (t, u) . ˚ ··· E0 (t, u) . ˚ 2 We recover C k bounds for Ψ via Sobolev We recover C k bounds for µ, Li from the eikonal equation Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Energy estimate (divergence theorem) region Σut Mt,u Cut t C0 Ψ ≡ 0 Σu0 E0 (t,u) z Z }| { z Z F0 (t,u) }| { P1 [∂Ψ, ∂Ψ] + P2 [∂Ψ, ∂Ψ] Cut Z Z = P1 [∂Ψ, ∂Ψ] + Q[∂Ψ, ∂Ψ] Σut Σu0 + Z Mt,u Mt,u 0 o z }| { n (g(Ψ) Ψ) (1 + µ)LΨ + 2R̆Ψ Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Energies and fluxes feature µ weights E0 (t, u) ∼ F0 (t, u) ∼ Z Σut µ|LΨ|2 + |R̆Ψ|2 + µ|∇ / Ψ|2 , Cut |LΨ|2 + µ|∇ / Ψ|2 Z Similarly, E1 (t, u) ∼ Z Σut µ|LZ Ψ|2 + |R̆Z Ψ|2 + µ|∇ / Z Ψ|2 , etc., where the Z are vectorfields depending on ∂u, Ψ, ∂Ψ Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Energies and fluxes feature µ weights E0 (t, u) ∼ F0 (t, u) ∼ Z Σut µ|LΨ|2 + |R̆Ψ|2 + µ|∇ / Ψ|2 , Cut |LΨ|2 + µ|∇ / Ψ|2 Z Similarly, E1 (t, u) ∼ Z Σut µ|LZ Ψ|2 + |R̆Z Ψ|2 + µ|∇ / Z Ψ|2 , etc., where the Z are vectorfields depending on ∂u, Ψ, ∂Ψ Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Top-order degeneracy Why is the top-order estimate so degenerate? Because the best a priori inequality we are able to obtain is: 2 E24 (t, u) ≤ ˚ +B B ∼ 10 Z t t 0 =0 ! ∂ sup ln µ E24 (t 0 , u) dt 0 + · · · ∂t Σt 0 The key difficult factor supΣt 0 ∂t∂ ln µ is connected to avoiding derivative loss in u. Caricature estimate: if µ were a function of only t, then Gronwall =⇒ E24 (t, u) . ˚ 2 µ−B + · · · The correct Gronwall-type estimate is hard to prove because µ = µ(t, u, ϑ) Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Top-order degeneracy Why is the top-order estimate so degenerate? Because the best a priori inequality we are able to obtain is: 2 E24 (t, u) ≤ ˚ +B B ∼ 10 Z t t 0 =0 ! ∂ sup ln µ E24 (t 0 , u) dt 0 + · · · ∂t Σt 0 The key difficult factor supΣt 0 ∂t∂ ln µ is connected to avoiding derivative loss in u. Caricature estimate: if µ were a function of only t, then Gronwall =⇒ E24 (t, u) . ˚ 2 µ−B + · · · The correct Gronwall-type estimate is hard to prove because µ = µ(t, u, ϑ) Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Top-order degeneracy Why is the top-order estimate so degenerate? Because the best a priori inequality we are able to obtain is: 2 E24 (t, u) ≤ ˚ +B B ∼ 10 Z t t 0 =0 ! ∂ sup ln µ E24 (t 0 , u) dt 0 + · · · ∂t Σt 0 The key difficult factor supΣt 0 ∂t∂ ln µ is connected to avoiding derivative loss in u. Caricature estimate: if µ were a function of only t, then Gronwall =⇒ E24 (t, u) . ˚ 2 µ−B + · · · The correct Gronwall-type estimate is hard to prove because µ = µ(t, u, ϑ) Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Open problems Important open problems: proving shock formation for systems with multiple characteristics Figure: System with two characteristics Cosmology (Euler-Einstein) Magnetohydrodynamics Elasticity Nonlinear electromagnetism Crystal optics Intro Previous work Main results Geometric formulation Estimates Energy hierarchy Latest preprints of the book and survey article are available at: http://math.mit.edu/ jspeck/publications.html