Partial Differential Equations III & V, Exercise Sheet 3: Solutions
Lecturer: Alpár R. Mészáros

  1. Existence of classical solutions. Consider the conservation law \[\begin{aligned} u_t + u^3 u_x = 0 \quad & \textrm{for } (x,t) \in \mathbb{R} \times (0,\infty), \\ u(x,0) = u_0(x) \quad & \textrm{for } x \in \mathbb{R}, \end{aligned}\] where \[u_0(x) = \left\{ \begin{array}{cl} 1 & x<0, \\ (1-x^2)^2 & 0 \le x \le 1, \\ 0 & x > 1. \end{array} \right.\] We have \(c(u)=u^3\) and \[c(u_0(s)) = \left\{ \begin{array}{cl} 1 & s<0, \\ (1-s^2)^6 & 0 \le s \le 1, \\ 0 & s > 1. \end{array} \right.\] Note that \(c(u_0(s))\) is decreasing for \(s \in (0,1)\). By Theorem 3.5, the conservation law has a classical solution \(u:\mathbb{R} \times [0,t_{\mathrm{c}}) \to \mathbb{R}\) where \[t_{\mathrm{c}} = \inf_{s \in (0,1)} \frac{-1}{c(u_0(s))_s} = \inf_{s \in (0,1)} \frac{1}{12 s (1-s^2)^5}.\] Note that \(1/(12 s (1-s^2)^5) \to \infty\) as \(s \to 0,1\). Therefore its minimum in \((0,1)\) is attained at a critical point. We compute \[\begin{aligned} \frac{d}{ds} \frac{1}{12 s (1-s^2)^5} & =\frac{d}{ds} (12 s(1-s^2)^5)^{-1} \\ & = -(12 s(1-s^2)^5)^{-2} (12(1-s^2)^5 -120s^2(1-s^2)^4) \\ & = \frac{-12 (1-s^2)^4(1-11s^2)}{(12 s(1-s^2)^5)^{2}}. \end{aligned}\] If follows that, for \(s \in (0,1)\), \[\frac{d}{ds}\frac{1}{12 s (1-s^2)^5} = 0 \quad \Longleftrightarrow \quad 1-11s^2 = 0 \quad \Longleftrightarrow \quad s^2 = \frac{1}{11}.\] Therefore the only critical point of \(1/(12 s (1-s^2)^5)\) in \((0,1)\) is \(s= 1/\sqrt{11}\), and this is its global minimum point. Its minimum value is \[t_{\mathrm{c}} = \left. \frac{1}{12 s (1-s^2)^5} \right|_{s=1/\sqrt{11}} = \frac{\sqrt{11}}{12(10/11)^5} = \boxed{\frac{11^5 \sqrt{11}}{10^5 12} \approx 0.445}\]

  2. Breakdown of classical solutions. Consider the conservation law \[\begin{aligned} u_t + u u_x = 0 \quad & \textrm{for } (x,t) \in \mathbb{R} \times (0,\infty), \\ u(x,0) = \frac{1}{1+x^2} \quad & \textrm{for } x \in \mathbb{R}. \end{aligned}\]

  3. Geometric interpretation of \(t_{\mathrm{c}}\): Crossing characteristics.

    (i) Let \(t>t_{\mathrm{c}}\) and \(x=s_{\mathrm{c}}+c(u_0(s_{\mathrm{c}}))t\). Define \(h(s,t)=s+c(u_0(s))t\). Note that \(h(s_{\mathrm{c}},t)=x\) by definition of \(x\). Recall that \[t_{\mathrm{c}} = \frac{-1}{c'(u_0(s_{\mathrm{c}}))u_0'(s_{\mathrm{c}})}.\] We have \[h_s(s,t) = 1+ c(u_0(s))_s t = 1+c'(u_0(s))u_0'(s)t.\] Therefore \[h_s(s_{\mathrm{c}},t) = 1+c'(u_0(s_{\mathrm{c}}))u_0'(s_{\mathrm{c}})t = 1 - \frac{t}{t_{\mathrm{c}}} < 1 - 1 = 0\] because \(t>t_{\mathrm{c}}\). Since \(h(s_{\mathrm{c}},t)=x\) and \(h_s(s_{\mathrm{c}}, t)<0\), then there exists \(s_1>s_{\mathrm{c}}\) such that \(h(s_1,t)<x\). But \(h(s,t) \to \infty\) as \(s \to \infty\). In particular, there exists \(s_2 > s_1\) with \(h(s_2,t)>x\). Since \(h\) is continuous, the Intermediate Value Theorem implies that there exists \(s_* \in (s_1,s_2)\) such that \(h(s_*,t)=x\). In conclusion, \[h(s_{\mathrm{c}},t) = x = h(s_*,t)\] and hence \((x,t)\) lies on the characteristic through \((s_{\mathrm{c}},0)\) and on the characteristic through \((s_*,0)\).

    (ii) Now, if we have two crossing characteristics \(x_1(t)=s_1+tc(u_0(s_1))\) and \(x_2(t)=s_2+tc(u_0(s_2))\), crossing at some point \((x,t)\), we have that \(x_1(t)=x_2(t) =x\). This means, after rearranging that \[s_1-s_2 = t\left(c(u_0(s_2))-c(u_0(s_1))\right).\] By the assumptions of the problem, we have that the mapping \(\mathbb{R}\ni s\mapsto c(u_0(s))\) is Lipschitz continuous. Actually, this is also Lipschitz continuous when one restricts the domain to \(I:=\{s\in\mathbb{R}: c'(u_0(s))u'_0(s)<0\}\), so define \[L_I:=\sup_{s\in I}|c'(u_0(s))u'_0(s)|.\] By the previous arguments, one must have \[|s_1-s_2| = t |c(u_0(s_2))-c(u_0(s_1))|\le t L_I |s_1-s_2|,\] which clearly implies that \(t\ge\frac{1}{L_I}\). Now, comparing this to the definition of \(t_c\), i.e. \[t_c:=\inf_{s\in I} - \frac{1}{c'(u_0(s))u_0'(s)} = \frac{1}{L_I},\] thus \(t\ge t_c\).

    It is very important to notice that in the previous argument we assumed that \(s_1\) and \(s_2\) are located in the same connected component of \(I\).

    Otherwise, let us suppose that \(I\) is disconnected and \(s_1\) and \(s_2\) are located in two different connected components of \(I\). Without loss of generality, we can suppose that \(s_1<s_2\). In this case, we can write again \[|s_1-s_2| = t |c(u_0(s_2))-c(u_0(s_1))| = t |c'(u_0(s_0))u'_0(s_0)| |s_1-s_2|,\] where in the second equation we have used the mean value property for derivatives and \(s_0\in (s_1,s_2)\). Now, if \(s_0\in I\), we are again done by the previous arguments.

    Otherwise, we have that \(s_0\notin I\) and so \(c'(u_0(s_0))u'_0(s_0)\ge 0\). In this case if \[0\le c'(u_0(s_0))u'_0(s_0)\le L_I,\] we are again done. So it remains to show that for these crossing characteristics with base points \(s_1,s_2\in I\), one cannot have that \[c'(u_0(s_0))u'_0(s_0)> L_I.\] Indeed, if this would be the case, it would mean that \(c\circ u_0\) can increase way faster than as it is decreasing. Why such a condition is possible, in such a scenario we would necessary need to have that if crossing is occurring after a base point \(s_0\) where the function increases a lot, in the mean value theorem one must have that \(c'(u_0(s_0))u'_0(s_0)<0\) or at least \(c'(u_0(s_0))u'_0(s_0)\le L_I\).

  4. The Fundamental Lemma of the Calculus of Variations. Let \(g: \Omega \to \mathbb{R}\) be a continuous function on an open set \(\Omega \subseteq \mathbb{R}^n\). Suppose that \[\begin{equation} \label{FLotCoV} \int_\Omega g(\boldsymbol{x}) \psi(\boldsymbol{x}) \, d \boldsymbol{x}= 0 \quad \textrm{for all } \, \psi \in C_c^\infty(\Omega). \end{equation}\] We show that \(g=0\). Assume for contradiction that \(g \ne 0\). Without loss of generality, assume that there exists a point \(\boldsymbol{p}\in \Omega\) such that \(g(\boldsymbol{p})>0\). Since \(g\) is continuous, then there exists an open set \(U \subset \Omega\) containing \(\boldsymbol{p}\) such that \(g(\boldsymbol{x})>0\) for all \(\boldsymbol{x}\in U\). Choose \(\varepsilon > 0\) such that \(B_\varepsilon(\boldsymbol{p}) \subset U\), where \(B_\varepsilon(\boldsymbol{p})\) is the open ball of radius \(\varepsilon\) centred at \(\boldsymbol{p}\). Choose any nonzero \(\varphi \in C_c^{\infty}(B_\varepsilon(\boldsymbol{p}))\) such \(\varphi \ge 0\), i.e., \(\varphi(\boldsymbol{x}) \ge 0\) for all \(\boldsymbol{x}\in B_\varepsilon(\boldsymbol{p})\). For example, we could take \[\varphi(\boldsymbol{x}) = \left\{ \begin{array}{cl} \exp \left( - \frac{1}{1-(2 | \boldsymbol{x}- \boldsymbol{p}| / \varepsilon) ^2 } \right) & \textrm{if } |\boldsymbol{x}-\boldsymbol{p}| < \varepsilon/2, \\ 0 & \textrm{if } |\boldsymbol{x}-\boldsymbol{p}| \ge \varepsilon/2. \end{array} \right.\] Then \[\int_\Omega g(\boldsymbol{x}) \varphi(\boldsymbol{x}) \, d \boldsymbol{x} = \int_{B_\varepsilon(\boldsymbol{p})} g(\boldsymbol{x}) \varphi(\boldsymbol{x}) \, d \boldsymbol{x}> 0,\] which contradicts equation \(\eqref{FLotCoV}\). Therefore \(g=0\), as required.

  5. Weak formulation of scalar conservation laws. In order to write down the weak formulation of the conservation law we need to compute a flux function \(f\), which satisfies \(f'(u)=c(u)=\ln|u|+1\). Note that \(f\) is only unique up to a constant. We can choose \[\begin{aligned} f(u) & = \int_0^u (\ln |s| + 1) \, ds = \int_0^u \left( \frac{ds}{ds} \ln |s| + 1 \right) \, ds \\ & = s \ln |s| \Big|_{0}^u - \int_0^u s \, \frac{d}{ds} \ln |s| \, ds + \int_0^u 1 \, ds = u \ln |u| - \int_0^u 1 \, ds + \int_0^u 1 \, ds \\ & = u \ln |u|. \end{aligned}\] Therefore \(u \in L^{\infty}(\mathbb{R} \times (0,\infty))\) is a weak solution of the conservation law if \[\boxed{ \int_{t=0}^\infty \int_{x=-\infty}^\infty [u(x,t) \varphi_t(x,t)+u(x,t) \ln |u(x,t)| \varphi_x (x,t)] \, dx dt + \int_{-\pi}^\pi (1+ \cos x) \varphi(x,0) \, dx = 0 }\] for all \(\varphi \in C_c^\infty(\mathbb{R} \times [0,\infty))\).

  6. Weak formulation of scalar conservation laws in several variables. We will use the following integration by parts formula, which can be proved in the same way as Lemma 3.15: Let \(u:\mathbb{R}^n \times [0,\infty) \to \mathbb{R}\), \(\boldsymbol{v}:\mathbb{R}^n \times [0,\infty) \to \mathbb{R}^n\) be continuously differentiable and let \(\varphi \in C_c^\infty(\mathbb{R}^n \times [0,\infty))\). Then \[\begin{gathered} \int_{0}^\infty \int_{\mathbb{R}^n} [u_t(\boldsymbol{x},t)+\mathrm{div} \, \boldsymbol{v}(\boldsymbol{x},t)] \varphi(\boldsymbol{x},t) \, d\boldsymbol{x}dt = \\ - \int_{\mathbb{R}^n} u(\boldsymbol{x},0) \varphi(\boldsymbol{x},0) \, d \boldsymbol{x} - \int_{0}^\infty \int_{\mathbb{R}^n} [u(\boldsymbol{x},t) \varphi_t(\boldsymbol{x},t) + \boldsymbol{v}(\boldsymbol{x},t) \cdot \nabla \varphi(\boldsymbol{x},t)] \, d\boldsymbol{x}dt. \end{gathered}\]

  7. The Rankine-Hugoniot condition. Consider Burger’s equation with initial data \[u_0(x) = \left\{ \begin{array}{cl} 2 & \textrm{if } x<0, \\ 1 & \textrm{if } 0<x<1, \\ 0 & \textrm{if } x>1. \end{array} \right.\]

  8. The Lax entropy condition. We need to find \(a,b,c,d>0\) so that the conservation law \(u_t+uu_x=0\) is satisfied piecewise, so that the Rankine-Hugoniot condition is satisfied across the shock \(x=-dt^2\), and so that the Lax entropy condition holds. Let \[u(x,t) = \left\{ \begin{array}{ll} -a (t + \sqrt{bx+ct^2}) & \textrm{if } x > -dt^2, \\ 0 & \textrm{if } x < -dt^2. \end{array} \right.\] Define \[r = r(x,t) = \sqrt{bx + ct^2}.\] Clearly \(u_t+uu_x=0\) is satisfied for \(x<-dt^2\). Now consider the case \(x>-dt^2\). Then \[\begin{aligned} u_t = -a - \frac{act}{r}, \qquad u_x = -\frac{ab}{2r}. \end{aligned}\] Let’s check when the conservation law is satisfied: \[\begin{aligned} u_t + u u_x = 0 \quad & \Longleftrightarrow \quad -a - \frac{act}{r} +\frac{a^2 b}{2r}(t+r) = 0 %\\ %& \Longleftrightarrow \quad %\frac{-2ar -2act + a^2b(t+r)}{2r} = 0 \\ & \Longleftrightarrow \quad \frac{-2ar -2act + a^2b(t+r)}{2r} = 0 \\ & \Longleftrightarrow \quad -2r -2ct + ab(t+r) = 0 \\ & \Longleftrightarrow \quad t(ab-2c)+r(ab-2)=0. \end{aligned}\] Since this must hold for all \(x,t\) with \(x>-dt^2\), we have \(ab-2c=0\) and \(ab-2=0\). Therefore \[\boxed{ab=2} \qquad \boxed {c=1}\] Now we check when the Rankine-Hugoniot condition is satisfied. We have \[\sigma(t)=-dt^2, \quad f(u)=\frac12 u^2, \quad u_l = 0, \quad u_r = -a(t+\sqrt{b\sigma(t)+ct^2}) = -at (1+\sqrt{c-bd}).\] Therefore the Rankine-Hugoniot condition holds if and only if \[\begin{aligned} \dot{\sigma} = \frac{[[f(u)]]}{[[u]]} & \quad \Longleftrightarrow \quad -2dt = \frac{\frac12 u_l^2 - \frac12 u_r^2}{u_l - u_r} \\ %& \quad \Longleftrightarrow \quad -2dt = \frac{\frac12 u_l - \frac12 u_r}{u_l - u_r} %\\ & \quad \Longleftrightarrow \quad -2dt = \frac{1}{2} (u_l + u_r) \\ & \quad \Longleftrightarrow \quad - 4 dt = -at(1+\sqrt{c-bd}) \\ & \quad \Longleftrightarrow \quad \boxed{ 4d = a(1+\sqrt{c-bd})} \end{aligned}\] Finally, since \(c(u)=u\), the Lax entropy condition holds if and only if \[c(u_r) < \dot{\sigma} < c(u_l) \quad \Longleftrightarrow \quad -at(1+\sqrt{c-bd}) < -2dt < 0 \quad \Longleftrightarrow \quad -4dt < -2dt < 0.\] This is satisfied for all \(t>0\) if \[\boxed{d>0}\]

    To check whether this solution is an entropy solution, we need to check whether is the solution uniformly bounded and is there \(C>0\) such that \[u(x+z,t)- u(x,t) \le C(1+1/t)z,\ \ \forall (x,t)\in\mathbb{R}\times(0,+\infty),\ \ \forall\ z>0.\] First, let us notice that the previously derived conditions on the constants imply that all these constants must be positive ones. Therefore, for each fixed time slice \(t\), we have that \(x\mapsto u(x,t)\) is non-increasing. Therefore, we have that \[u(x+z,t)- u(x,t) \le 0,\ \ \forall (x,t)\in\mathbb{R}\times(0,+\infty),\ \ \forall\ z>0.\] However, it is clear to see to this solution is not uniformly bounded, therefore by our definition it is not an entropy solution.

  9. The Lax entropy condition again.

  10. Rarefaction waves and shocks.

  11. Watching paint dry.

  12. Traffic flow.

  13. Strong solutions. The proof is somewhat similar to the derivation of the Rankine-Hugoniot condtion (or rather to the proof of the converse of the Rankine-Hugoniot condition). Let \(V = \mathbb{R} \times (0,\infty)\) and let \(\varphi \in C_c^\infty(V)\). Then \[\begin{aligned} & \int_V [u \varphi_t + f(u) \varphi_x] \, dx dt + \int_{-\infty}^\infty u_0(x) \varphi(x,0) \, dx \\ & = \int_{V_l} [u \varphi_t + f(u) \varphi_x] \, dx dt + \int_{V_r} [u \varphi_t + f(u) \varphi_x] \, dx dt + \int_{-\infty}^\infty u_0(x) \varphi(x,0) \, dx \\ & = \int_{V_l} (f(u),u) \cdot \nabla_{(x,t)} \varphi \, dx dt + \int_{V_r} (f(u),u) \cdot \nabla_{(x,t)} \varphi \, dx dt + \int_{-\infty}^\infty u_0(x) \varphi(x,0) \, dx \\ & = \int_{\partial V_l} (f(u),u) \varphi \cdot \boldsymbol{n}_l \, dL - \int_{V_l} \mathrm{div}_{(x,t)} (f(u),u) \varphi \, dx dt +\int_{\partial V_r} (f(u),u) \varphi \cdot \boldsymbol{n}_r \, dL - \int_{V_r} \mathrm{div}_{(x,t)} (f(u),u) \varphi \, dx dt \\ & \phantom{=} \; \; + \int_{-\infty}^\infty u_0(x) \varphi(x,0) \, dx \end{aligned}\] where \(\boldsymbol{n}_l\) and \(\boldsymbol{n}_r\) are the unit outward-pointing normal vectors to \(\partial V_l\), \(\partial V_r\). Observe that \[\partial V_l = \{ (x,0) : x \le g(0) \} \, \cup \, \gamma, \quad \partial V_r = \{ (x,0) : x \ge g(0) \} \, \cup \, \gamma.\] Moreover \(\boldsymbol{n}_r = - \boldsymbol{n}_l \textrm{ on } \gamma\), \(\boldsymbol{n}_l = -(0,1)\) on \(\partial V_l \setminus \gamma\), \(\boldsymbol{n}_r = -(0,1)\) on \(\partial V_r \setminus \gamma\). Continuing the calculation gives \[\begin{aligned} & \int_V [u \varphi_t + f(u) \varphi_x] \, dx dt + \int_{-\infty}^\infty u_0(x) \varphi(x,0) \, dx \\ & = - \int_{-\infty}^{g(0)} (f(u),u) \varphi \big|_{t=0} \cdot (0,1) \, dx + \int_{\gamma} (f(u),u) \varphi \cdot \boldsymbol{n}_l \, dL - \int_{V_l} (\underbrace{u_t + f(u)_x}_{=0} ) \varphi \, dx dt \\ & \phantom{=} \; \; - \int_{g(0)}^\infty (f(u),u) \varphi \big|_{t=0} \cdot (0,1) \, dx + \int_{\gamma} (f(u),u) \varphi \cdot \boldsymbol{n}_r \, dL - \int_{V_r} (\underbrace{u_t + f(u)_x}_{=0} ) \varphi \, dx dt + \int_{-\infty}^\infty u_0(x) \varphi(x,0) \, dx \\ & = - \int_{-\infty}^{g(0)} u(x,0) \varphi(x,0) \, dx + \int_{\gamma} (f(u),u) \varphi \cdot \boldsymbol{n}_l \, dL- \int_{V_l} (\underbrace{u_t + f(u)_x}_{=0} ) \varphi \, dx dt \\ & \phantom{=} \; \; - \int_{g(0)}^\infty u(x,0) \varphi(x,0) \, dx + \int_{\gamma} (f(u),u) \varphi \cdot \underbrace{\boldsymbol{n}_r}_{=-\boldsymbol{n}_l} \, dL- \int_{V_r} (\underbrace{u_t + f(u)_x}_{=0} ) \varphi \, dx dt + \int_{-\infty}^\infty u_0(x) \varphi(x,0) \, dx \\ & = \int_{\infty}^\infty (\underbrace{u_0(x) - u(x,0)}_{=0}) \varphi(x,0) \, dx \\ & = 0. \end{aligned}\] This holds for all \(\varphi \in C_c^\infty(V)\). Therefore \(u\) is a weak solution of the conservation law, as required.