betaFeedback

WorksMathematics

On the Minimum Speed for a Fire Blocking Problem

Abstract

We consider a dynamic blocking problem: a fire, starting on a bounded open set R0⊂R2R_0\subset\mathbb R^2, propagates in the plane with unit speed in all directions. To block it, a barrier can be constructed in real time, at speed λ>0λ>0. A long standing conjecture by the first author states that the fire can be entirely surrounded by the barrier, in finite time, if and only if λ>2λ>2. The present paper improves on earlier lower bounds. Namely, if λ≤Λ=1.62621170228…λ\leq Λ=1.62621170228\ldots we show that the fire cannot be blocked. We also prove that, if λ≤2λ\leq 2, the fire cannot be blocked by constructing a barrier which is a simple closed curve. More precisely, if the barrier does not contain purely delaying arcs, i.e.~arcs which are eventually surrounded by the fire on both sides, then for λ<2λ<2 the region burned by the fire at time t>0t>0 has area A(t)>(2−λ)t2/4A(t)>(2-λ)t^2/4.

Introduction

Aim of this note is to analyze the dynamic blocking problem introduced in [7], originally motivated by the control of wild fires or of the spatial spreading of a contaminating agent.

At each time t≥0t \ge0, we denote by Rt⊂R2R_t \subset\mathbb{R}^2 the burned region. In the absence of control, we assume that the set R(t)R(t) grows uniformly in all directions, namely

Rt=˙B(R0,t)=˙{x∈R2:d(x,R0)≤t}.R_t \mathrel{\dot=} B(R_0,t) \mathrel{\dot=} \left\{x \in\mathbb{R}^2 : d(x,R_0) \le t\right\}.

Here R0⊂R2R_0 \subset\mathbb{R}^2 is a fixed (nonempty) bounded open set, describing the region burned by the fire at the initial time t=0t = 0. In our model, the spreading of the fire can be controlled by constructing barriers. In mathematical terms, we thus assume that the controller can construct a one-dimensional rectifiable curve γ\gamma which blocks the spreading of the fire. Calling γ(t)⊂R2\gamma(t) \subset\mathbb{R}^2 the portion of the wall constructed within time t≥0t \ge0, and denoting by H1\mathcal{H}^1 the one-dimensional Hausdorff measure, we assume:

(A1) For every t2>t1≥0t_2 > t_1 \ge0 one has γ(t1)⊆γ(t2)\gamma(t_1) \subseteq\gamma(t_2).

(A2) For every t≥0t \ge0, the total length of the barrier satisfies

H1(γ(t))≤λt.(1)\mathcal{H}^1(\gamma(t)) \le\lambda t. \tag*{(1)}

By (A1), after a portion of the barrier has been constructed, it remains in place forever and cannot be moved to any other location. The constant λ>0\lambda> 0 in (1) is the speed at which the wall is constructed. A barrier γ(⋅)\gamma(\cdot) satisfying (A1)–(A2) will be called a λ\lambda-admissible strategy. By constructing barriers, the burned set is reduced. Indeed, this set is

Rtγ≔{x(t);x(⋅) is absolutely continuous,x(0)∈R0,∣x˙(τ)∣≤1 for a.e. τ∈[0,t],x(τ)∉γ(τ) for all τ∈[0,t]}.(2)R_t^\gamma\coloneqq\left\{ \begin{aligned} &x(t); \quad x(\cdot)\text{ is absolutely continuous},\quad x(0) \in R_0,\\ &|\dot{x}(\tau)| \le1\text{ for a.e. }\tau\in[0,t],\quad x(\tau) \notin\gamma(\tau)\text{ for all }\tau\in[0,t] \end{aligned} \right\}. \tag*{(2)}

In the above setting, following [7] we consider

(BP) Blocking Problem. Given an initial bounded open set R0⊂R2R_0 \subset\mathbb{R}^2, find an admissible strategy t↦γ(t)t \mapsto\gamma(t) such that the corresponding reachable sets RtγR_t^\gamma remain uniformly bounded.

In other words, we seek a strategy such that

Rtγ⊆BR≔{x∈R2;∣x∣<R}for all t≥0,R_t^\gamma\subseteq B_R \coloneqq\{x \in\mathbb{R}^2; |x| < R\}\qquad\text{for all }t \ge0,

for some radius RR sufficiently large. A key question that still remains open is

(Q) For which construction speeds λ\lambda can the fire be blocked?

By rescaling argument, it is clear that the answer does not depend on the initial set R0R_0. Indeed, if γ(⋅)\gamma(\cdot) is a λ\lambda-admissible strategy that blocks a fire starting from the set R0R_0, then for every c>0c > 0 the rescaled strategy

γ~(t)=cγ(t/c)\widetilde{\gamma}(t) = c\gamma(t/c)

is a λ\lambda-admissible strategy that blocks a fire initially burning on the set cR0cR_0. W.l.o.g. one can thus assume that R0={x∈R2;∣x∣<1}R_0 = \{x \in\mathbb{R}^2; |x| < 1\} is the unit disc.

A long standing conjecture by the first author is that the fire can be blocked only if λ>2\lambda> 2. In this direction, the following results are known.

  1. When λ>2\lambda> 2 the fire can be blocked. When λ≤1\lambda\le1 it cannot be blocked [7].

  1. If the fire propagation is restricted to a half plane, then it can be blocked if and only if λ>1\lambda> 1, see [11].

  1. A spiral-like strategy which blocks the fire can be constructed only if λ>λ∗≔2.614430844…\lambda> \lambda^* \coloneqq2.614430844\ldots [6, 8].

(4) If λ≤3/2\lambda\le3/2, the fire cannot be blocked [19].

In the present note we point out a very simple argument showing that, when λ≤π/2\lambda\le\pi/2, the fire cannot be blocked. Indeed, calling A(t)A(t) the area of the burned region at time t≥0t \ge0, by the isoperimetric inequality the boundary of the burned set has length ≥2πA(t)\ge2\sqrt{\pi A(t)}. Of this boundary, at most a portion of length λt\lambda t can be covered by walls. On the remaining portion FtF_t the fire expands freely (see Figure 1, left). This yields the lower bound

Figure 1. Left: by the isoperimetric inequality, at any time t>0t > 0 the portion FtF_t of the boundary where the fire expands has length ≥2πA(t)−λt\ge2\sqrt{\pi A(t)}-\lambda t. Right: since for τ>t\tau> t the barrier γ(τ)\gamma(\tau) must contain γ(t)\gamma(t), at one of the two times tt or τ\tau the inequality (3) must be strict. For this reason, the bound in (10) can be slightly improved.

ddtA(t)≥2πA(t)−λt.(3)\frac{d}{dt} A(t) \ge2\sqrt{\pi A(t)}-\lambda t. \tag*{(3)}

When λ≤π/2\lambda\le\pi/2, this differential inequality yields A(t)≥ct2A(t) \ge ct^2 for some c>0c > 0 and all t≥0t \ge0, showing that the fire cannot be blocked.

This lower bound on the blocking speed λ\lambda can be improved by observing that, for any τ>t\tau> t, the barrier γ(τ)\gamma(\tau) must contain γ(t)\gamma(t) (see Figure 1, right). Hence at one of the two times tt or τ\tau the inequality (1.3) must be strict. A lower bound on the sum A′(t)+A′(τ)A'(t) + A'(\tau) can now be obtained by looking at a double-bubble minimization problem, for the weighted sum of the perimeters of two sets with areas A(t)A(t) and A(τ)−A(t)A(\tau)-A(t).

Given a time T>0T > 0, for any couple of times t∈[0,T/2]t \in[0,T/2] and τ=t+T/2\tau= t+T/2 the bounds known for the double-bubble problem [1, 16, 18] provide a sharper estimate on the lengths of the free boundaries FtF_t and FτF_\tau where the fire expands. In particular, when λ≤Λ:=1.62621170228…\lambda\le\Lambda\mathrel{:=} 1.62621170228\ldots an integration yields

A(T)=A(0)+∫0T/2[A′(t)+A′(t+T/2)] dt≥c∗T2A(T)=A(0)+\int_{0}^{T/2}\left[A'(t)+A'(t+T/2)\right]\,dt \ge c_*T^2

for some c∗>0c_* > 0 and all T≥0T \ge0. Again, this shows that the fire cannot be blocked.

The last section of the paper is concerned with the case where the entire barrier Γ=⋃t>0γ(t)\Gamma=\bigcup_{t>0}\gamma(t) is a Jordan curve, i.e. a simple closed curve. In such case we show that the fire can be blocked if and only if λ>2\lambda> 2. If λ<2\lambda< 2 and the barrier does not contain purely delaying arcs, i.e. arcs that are eventually surrounded by the fire from both sides, we prove that the area of the burned set satisfies A(t)≥(2−λ)t2/4A(t) \ge(2-\lambda)t^2/4. This indicates that, if a λ\lambda-admissible blocking strategy were to exist for some λ<2\lambda< 2, it would have to contain some barriers whose only purpose is to delay the spreading of the fire. As time increases, these barriers eventually lie in the interior of the closure of the burned set.

For the definition and properties of rectifiable sets we refer to [3, 14, 17]. Related problems of fire confinement have been studied in [5, 15]. In connection with a cost functional accounting for the value of the burned region plus the cost of constructing the barrier, the existence of optimal barriers was proved in [9, 13]. Necessary conditions for optimality, describing each arc of the optimal barrier as the solution to a suitable ODE, were derived in [7, 12, 20].

Preliminary results on the minimum time function

Throughout this section we assume that R0R_0 is a bounded open set, whose boundary has finite length: H1(∂R0)<+∞\mathcal{H}^1(\partial R_0)<+\infty.

Consider the minimum time function associated with an admissible strategy γ\gamma:

uγ(x)≔inf⁡{t>0; x∈Rtγ},(4)u^\gamma(x) \coloneqq\inf\{t > 0;\ x \in R_t^\gamma\}, \tag*{(4)}

with the convention inf⁡∅=+∞\inf\varnothing= +\infty.

Following [3, 14], by a rectifiable set we mean a set Γ⊂R2\Gamma\subset\mathbb{R}^2 which can be decomposed as Γ=Γ0∪Γ1\Gamma= \Gamma_0 \cup\Gamma_1, where H1(Γ0)=0\mathcal{H}^1(\Gamma_0) = 0, H1(Γ1)<∞\mathcal{H}^1(\Gamma_1) < \infty, and Γ1\Gamma_1 is contained in a countable union of Lipschitz continuous curves. We say that the rectifiable set Γ\Gamma is complete if it contains all of its points of positive upper density:

lim sup⁡r→0+H1(B(x,r)∩Γ)r>0⟹x∈Γ.\limsup_{r \to0^+} \frac{\mathcal{H}^1(B(x,r) \cap\Gamma)}{r} > 0 \quad\Longrightarrow\quad x \in\Gamma.

We recall that, if Σ\Sigma is any rectifiable set (possibly not complete), the closure Σ‾\overline{\Sigma} may well be the entire plane R2\mathbb{R}^2. However, the union

Σ∗≔Σ∪{x∈R2; lim sup⁡r→0+H1(B(x,r)∩Σ)r>0}\Sigma^* \coloneqq\Sigma\cup\left\{x \in\mathbb{R}^2;\ \limsup_{r \to0^+} \frac{\mathcal{H}^1(B(x,r) \cap\Sigma)}{r} > 0\right\}

is a complete rectifiable set, with H1(Σ∗)=H1(Σ)\mathcal{H}^1(\Sigma^*) = \mathcal{H}^1(\Sigma). Throughout the following, we refer to Σ∗\Sigma^* as the completion of the set Σ\Sigma.

Following [9, 13], an admissible strategy is called complete if, for every t≥0t \ge0,

γ(t)=⋂s>tγ(s),\gamma(t) = \bigcap_{s>t} \gamma(s),

and moreover γ(t)\gamma(t) contains every point x∈R2x \in\mathbb{R}^2 satisfying

lim sup⁡r↓0H1(γ(t)∩Br(x))r>0.\limsup_{r \downarrow0} \frac{\mathcal{H}^1(\gamma(t) \cap B_r(x))}{r} > 0.

We recall that by [13 Theorem 1] the minimum-time function v=uγ:R2→[0,+∞]v=u^\gamma:\mathbb{R}^2\to[0,+\infty] corresponding to an admissible strategy γ(⋅)\gamma(\cdot) satisfies the following regularity properties. For every s>0s>0, the truncation vs=min⁡{v,s}v_s=\min\{v,s\} satisfies

vs∈SBVloc(R2),vs=0a.e. on R0,v_s\in SBV_{\mathrm{loc}}(\mathbb{R}^2),\qquad v_s=0\quad\text{a.e. on }R_0,

and moreover the jump set of vsv_s and the absolutely continuous part of its gradient satisfy

H1(Jvs∖γ(s))=0,∣∇avs∣≤1a.e.\mathcal{H}^1(J_{v_s}\setminus\gamma(s))=0,\qquad\lvert\nabla^a v_s\rvert\le1\quad\text{a.e.}

We next relate the growth of the burned area to the free front. Here and in the sequel, ∣Ω∣\lvert\Omega\rvert denotes the Lebesgue measure of a set Ω⊂R2\Omega\subset\mathbb{R}^2, while ∂∗Ω\partial^*\Omega denotes its reduced boundary (see [3, 17] for precise definitions).

Proposition 2.1 Let γ\gamma be a complete, admissible strategy and set A(t)=∣Rtγ∣A(t)=\lvert R_t^\gamma\rvert, with A(0)=∣R0∣A(0)=\lvert R_0\rvert. Then AA is nondecreasing and RtγR_t^\gamma has finite perimeter for almost every t>0t>0. For such times, define

Wt≔γ(t)∩∂∗Rtγ,Ft≔∂∗Rtγ∖γ(t).(5)W_t\coloneqq\gamma(t)\cap\partial^*R_t^\gamma,\qquad F_t\coloneqq\partial^*R_t^\gamma\setminus\gamma(t). \tag*{(5)}

Then

∂∗Rtγ=Wt∪Ft,Wt∩Ft=∅.\partial^*R_t^\gamma=W_t\cup F_t,\qquad W_t\cap F_t=\varnothing.

Moreover, for all τ1<τ2\tau_1<\tau_2 there holds

A(τ2)≥A(τ1)+∫τ1τ2H1(Fs) ds.(6)A(\tau_2)\ge A(\tau_1)+\int_{\tau_1}^{\tau_2}\mathcal{H}^1(F_s)\,ds. \tag*{(6)}

Proof. Fix τ>τ2\tau>\tau_2. By the coarea formula applied for the absolutely continuous part of the gradient of uτγu_\tau^\gamma, we have

A(τ2)−A(τ1)≥∣∇auτγ∣({τ1<uτγ≤τ2})=∫τ1τ2H1(∂∗{uτγ≤s}∖Juτγ) ds.A(\tau_2)-A(\tau_1)\ge\lvert\nabla^a u_\tau^\gamma\rvert\bigl(\{\tau_1<u_\tau^\gamma\le\tau_2\}\bigr)=\int_{\tau_1}^{\tau_2}\mathcal{H}^1\bigl(\partial^*\{u_\tau^\gamma\le s\}\setminus J_{u_\tau^\gamma}\bigr)\,ds.

Next, we claim that, for almost every s<τs<\tau, up to H1\mathcal{H}^1-negligible sets,

Juτγ∩∂∗{uτγ≤s}⊆Jusγ⊆γ(s).(7)J_{u_\tau^\gamma}\cap\partial^*\{u_\tau^\gamma\le s\}\subseteq J_{u_s^\gamma}\subseteq\gamma(s). \tag*{(7)}

Indeed, for almost every time ss, we claim that the following holds. For H1\mathcal{H}^1-a.e. x∈Juτγ∩∂∗{uτγ≤s}x\in J_{u_\tau^\gamma}\cap\partial^*\{u_\tau^\gamma\le s\}, the left and right traces (uτγ)±(x)(u_\tau^\gamma)^\pm(x) satisfy

(uτγ)−(x)<s<(uτγ)+(x).(u_\tau^\gamma)^-(x)<s<(u_\tau^\gamma)^+(x).

This is obtained using Fubini’s theorem:

0=∫Juτγ∫R1{uτγ±(x)=s}(x) ds dH1(x)=∫R∫Juτγ1{uτγ±(x)=s}(x) dH1(x) ds=∫RH1({x∈Juτγ∣uτγ±(x)=s}) ds.(8)\begin{aligned} 0=\int_{J_{u_\tau^\gamma}}\int_{\mathbb{R}}\mathbf{1}_{\{u_\tau^\gamma{}^\pm(x)=s\}}(x)\,ds\,d\mathcal{H}^1(x) \\ &=\int_{\mathbb{R}}\int_{J_{u_\tau^\gamma}}\mathbf{1}_{\{u_\tau^\gamma{}^\pm(x)=s\}}(x)\,d\mathcal{H}^1(x)\,ds \\ &=\int_{\mathbb{R}}\mathcal{H}^1\bigl(\{x\in J_{u_\tau^\gamma}\mid u_\tau^\gamma{}^\pm(x)=s\}\bigr)\,ds. \tag*{(8)} \end{aligned}

Therefore, for ss outside a negligible set,

(uτγ)±(x)≠sfor H1-a.e. point x∈Juτγ.(9)(u_{\tau}^{\gamma})^{\pm}(x) \ne s \qquad\text{for } \mathcal{H}^{1}\text{-a.e. point }x \in J_{u_{\tau}^{\gamma}}. \tag*{(9)}

If a point xx additionally belongs to Juτγ∩∂∗{uτγ≤s}J_{u_{\tau}^{\gamma}} \cap\partial^{*}\{u_{\tau}^{\gamma} \le s\}, then this means that the upper trace is greater than ss and the lower trace is less than ss, i.e. (uτγ)+(x)≥s≥(uτγ)−(x)(u_{\tau}^{\gamma})^{+}(x) \ge s \ge(u_{\tau}^{\gamma})^{-}(x), for H1\mathcal{H}^{1}-a.e. point in this intersection. Together with (2.6) this proves the claim. Therefore, for such xx, we immediately obtain x∈Jusγx \in J_{u_{s}^{\gamma}}. The second inclusion in (2.4) follows by the assumption that γ(⋅)\gamma(\cdot) is complete.

Consequently,

A(τ2)−A(τ1)≥∫τ1τ2H1(∂∗{uτγ≤s}∖γ(s)) ds=∫τ1τ2H1(Fs) ds.A(\tau_{2}) - A(\tau_{1}) \ge\int_{\tau_{1}}^{\tau_{2}} \mathcal{H}^{1}\bigl(\partial^{*}\{u_{\tau}^{\gamma} \le s\} \setminus\gamma(s)\bigr)\,ds = \int_{\tau_{1}}^{\tau_{2}} \mathcal{H}^{1}(F_{s})\,ds.

Remark 2.1 A more careful proof would show that (2.3) holds with equality. However, this is not needed for our present purposes.

An elementary lower bound on the blocking speed

In this section we provide a first, elementary lower bound for the minimum speed blocking the fire. We recall that A(t)=∣Rtγ∣A(t) = |R_{t}^{\gamma}| is the area of the burned region at time t>0t > 0, given an admissible strategy γ\gamma. We assume here that the initial set R0=B1R_{0} = B_{1} is the unit ball, so that A(0)=πA(0) = \pi.

Theorem 3.1 Let γ\gamma be any admissible strategy. In the above setting, if 0≤λ≤π/20 \le\lambda\le\pi/2, then

A(t)≥π(1+aλt)2,whereaλ≔1+1−2λ/π2.A(t) \ge\pi(1+a_{\lambda}t)^{2}, \qquad\text{where} \qquad a_{\lambda} \coloneqq\frac{1+\sqrt{1-2\lambda/\pi}}{2}.

In particular, when λ=π/2\lambda= \pi/2 one has

A(t)≥π(1+t2)2,(10)A(t) \ge\pi\left(1+\frac{t}{2}\right)^{2}, \tag*{(10)}

hence the fire cannot be contained.

Proof. By Proposition 2.1, there holds

A′(t)≥H1(Ft)(11)A'(t) \ge\mathcal{H}^{1}(F_{t}) \tag*{(11)}

for a.e. t>0t > 0. On the other hand, the barrier constructed up to time tt has length at most H1(γ(t))≤λt\mathcal{H}^{1}(\gamma(t)) \le\lambda t. Therefore

H1(∂∗Rtγ)=H1(Ft)+H1(Wt)≤A′(t)+λt.\mathcal{H}^{1}(\partial^{*}R_{t}^{\gamma}) = \mathcal{H}^{1}(F_{t}) + \mathcal{H}^{1}(W_{t}) \le A'(t) + \lambda t.

The planar isoperimetric inequality yields

A′(t)≥H1(∂∗Rtγ)−λt≥2πA(t)−λt.(12)A'(t) \ge\mathcal{H}^{1}(\partial^{*}R_{t}^{\gamma}) - \lambda t \ge2\sqrt{\pi A(t)} - \lambda t. \tag*{(12)}

Now set

Q(t)≔π(1+at)2Q(t) \coloneqq\pi(1+at)^2

and choose a∈[1/2,1]a \in[1/2,1] so that

2πa(1−a)=λ.(13)2\pi a(1-a)=\lambda. \tag*{(13)}

Such a number aa exists precisely when λ≤π/2\lambda\le\pi/2. Indeed

max⁡0≤a≤12πa(1−a)=π2.\max_{0\le a\le1} 2\pi a(1-a)=\frac{\pi}{2}.

Moreover, by (13)

2πQ(t)−λt−Q′(t)=2π(1−a)(1+at)−λt=2π(1−a)≥0,2\sqrt{\pi Q(t)}-\lambda t-Q'(t)=2\pi(1-a)(1+at)-\lambda t =2\pi(1-a)\ge0,

Therefore

Q′(t)≤2πQ(t)−λt.Q'(t)\le2\sqrt{\pi Q(t)}-\lambda t.

We now observe Q(0)=A(0)=πQ(0)=A(0)=\pi. Moreover, at any time tt where A(t)=Q(t)A(t)=Q(t), by (12) one has

A′(t)≥2πQ(t)−λt≥Q′(t).(14)A'(t)\ge2\sqrt{\pi Q(t)}-\lambda t\ge Q'(t). \tag*{(14)}

A standard comparison argument now yields

A(t)≥Q(t)for all t≥0.A(t)\ge Q(t)\qquad\text{for all }t\ge0.

At the critical value λ=π/2\lambda=\pi/2 the equation (13) has the double root a=1/2a=1/2. Hence

A(t)≥π(1+t2)2⟶∞,A(t)\ge\pi\left(1+\frac{t}{2}\right)^2\longrightarrow\infty,

showing that the fire cannot be blocked. Of course, the same conclusion holds for every λ<π/2\lambda<\pi/2.

Improving the lower bound

The earlier inequality (12) did not use the fact that for different times tt the sets γ(t)\gamma(t) belong to the same growing barrier. Relying on that additional information, the estimate can be somewhat improved.

Recall the definition of the active wall at time tt, that is

Wt≔γ(t)∩∂∗Rtγ.W_t\coloneqq\gamma(t)\cap\partial^*R_t^\gamma.

Notice that the remaining portion of the barrier γ(t)∖Wt\gamma(t)\setminus W_t consists of points where (see Figure 2)

Figure 2. Left: the solution to the double-bubble problem provides a lower bound on the lengths of the portions FtF_t, FτF_\tau of the boundaries where the fire expands freely, at the two times t<τt<\tau. Right: the portion of the barrier γ(t)\gamma(t) between the points PP and QQ is the active wall WtW_t at time tt. The remaining portions do not restrict the instantaneous fire propagation, because either they have not yet been reached by the fire, or they are already fully surrounded by it.

  • either the barrier has not yet been touched by the fire,

  • or the fire has already surrounded the barrier on both sides.

Fix two times where Proposition 2.1 applies, and consider the set

∂∗Raγ∩∂∗Rbγ=Wa∩Wb.\partial^* R_a^\gamma\cap\partial^* R_b^\gamma= W_a \cap W_b.

Note that these are the points which lie on an active wall at time aa, which remains active until time bb.

Introduce the function

Q(a,b):=H1(∂∗Raγ)+H1(∂∗Rbγ)−(2−2)H1(∂∗Raγ∩∂∗Rbγ).(15)Q(a,b) := \mathcal{H}^1(\partial^* R_a^\gamma) + \mathcal{H}^1(\partial^* R_b^\gamma) - (2-\sqrt{2})\mathcal{H}^1(\partial^* R_a^\gamma\cap\partial^* R_b^\gamma). \tag*{(15)}

The sets

Raγ,Rbγ∖Raγ,R2∖RbγR_a^\gamma,\qquad R_b^\gamma\setminus R_a^\gamma,\qquad\mathbb{R}^2 \setminus R_b^\gamma

form a partition of R2\mathbb{R}^2. In Q(a,b)Q(a,b) the interfaces between consecutive phases have weight 11, while the interface separating RaγR_a^\gamma directly from the exterior has weight 2\sqrt{2}.

In the following, recalling (2.2) we define

F(t):=H1(Ft)F(t) := \mathcal{H}^1(F_t)

to be the length of the advancing fire front, at time tt.

Proposition 4.1 There is a set N⊂R+N \subset\mathbb{R}^+ such that for every a,b∈R+∖Na,b \in\mathbb{R}^+ \setminus N, there holds

Q(a,b)≤F(a)+F(b)+λ[b+(2−1)a].(16)Q(a,b) \leq F(a) + F(b) + \lambda\left[b+(\sqrt{2}-1)a\right]. \tag*{(16)}

Proof. There holds

Q(a,b)−F(a)−F(b)=H1(Wa)+H1(Wb)−(2−2)H1(Wa∩Wb)=H1(Wa∪Wb)+(2−1)H1(Wa∩Wb).\begin{aligned} Q(a,b)-F(a)-F(b) &=\mathcal{H}^1(W_a)+\mathcal{H}^1(W_b)-(2-\sqrt{2})\mathcal{H}^1(W_a\cap W_b)\\ &=\mathcal{H}^1(W_a\cup W_b)+(\sqrt{2}-1)\mathcal{H}^1(W_a\cap W_b). \end{aligned}

Since the barrier is increasing in time,

Wa∪Wb⊆γ(b),Wa∩Wb⊆γ(a).W_a\cup W_b\subseteq\gamma(b),\qquad W_a\cap W_b\subseteq\gamma(a).

Using (1.1) gives (4.2).

Double-bubble variational problems

We consider the following optimization problem for sets Ω1,Ω2⊂R2\Omega_1,\Omega_2 \subset\mathbb{R}^2 of finite perimeter in the plane. As a convention, in the following let Ω0:=R2∖(Ω1∪Ω2)\Omega_0 := \mathbb{R}^2 \setminus(\Omega_1 \cup\Omega_2), and let

Γi,j:=∂∗Ωi∩∂∗Ωj.\Gamma_{i,j} := \partial^*\Omega_i \cap\partial^*\Omega_j.

Here Γi,j\Gamma_{i,j} are disjoint up to H1\mathcal{H}^1 negligible sets.

(DBP) (Double Bubble Problem). Given m,n≥0m,n \ge0, find two disjoint regions Ω1,Ω2⊂R2\Omega_1,\Omega_2 \subset\mathbb{R}^2 with areas m,nm,n respectively, which minimize the weighted sum of perimeters:

H1(Γ12)+H1(Γ20)+2⋅H1(Γ10).(17)\mathcal{H}^1(\Gamma_{12})+\mathcal{H}^1(\Gamma_{20})+\sqrt{2}\cdot\mathcal{H}^1(\Gamma_{10}). \tag*{(17)}

The weighted double-bubble theorem of Lawlor [16] identifies the minimizer with the standard weighted double bubble made of three circular arcs, as shown in Figure 3. The minimum cost D(m,n)\mathcal{D}(m,n) has the following properties:

Figure 3. The optimal shape of the double bubble.

D(cm,cn)=cD(m,n),for all c>0,(18)\mathcal{D}(cm,cn)=\sqrt{c}\mathcal{D}(m,n), \qquad\text{for all } c>0, \tag*{(18)}
D(0,n)=2πn,D(m,0)=22πm.(19)\mathcal{D}(0,n)=2\sqrt{\pi n}, \qquad\mathcal{D}(m,0)=2\sqrt{2\pi m}. \tag*{(19)}

Remark 4.1 Applying the definition of D\mathcal{D} to the two sets Ω1=Raγ\Omega_1=R_a^\gamma and Ω2=Rbγ∖Raγ\Omega_2=R_b^\gamma\setminus R_a^\gamma, one immediately obtains

Q(a,b)≥D(A(a),A(b)−A(a)).(20)Q(a,b)\ge\mathcal{D}(A(a),A(b)-A(a)). \tag*{(20)}

The following monotonicity property will also be useful.

Lemma 4.1 Let D(m,n)\mathcal{D}(m,n) be the minimum cost for the problem (DBP) with areas m,n>0m,n>0. Let v0>0v_0>0 and 0≤m0≤v0/40\le m_0\le v_0/4. Then for all 0≤m≤v0\le m\le v there holds

D(m,v−m)≥D(m0,v0−m0)for all 0<v0≤v,0≤m0≤min⁡{m,v04}.(21)\mathcal{D}(m,v-m)\ge\mathcal{D}(m_0,v_0-m_0)\qquad\text{for all }0<v_0\le v,\quad0\le m_0\le\min\left\{m,\frac{v_0}{4}\right\}. \tag*{(21)}

Proof. Set

G(m,v):=D(m,v−m),0≤m≤v.G(m,v):=\mathcal{D}(m,v-m),\qquad0\le m\le v.

Here mm is the area of the first chamber and vv is the total enclosed area. Increasing vv with mm fixed adds area to the second chamber, whereas increasing mm with vv fixed transfers area from the second chamber to the first.

We shall prove that, for fixed vv, the minimum of GG on [m0,v][m_0,v] is attained at one of its endpoints, and then compare both endpoint values with G(m0,v0)G(m_0,v_0).

We first recall the junction condition for a weighted double bubble; see [16 Section 6]. At each junction, let T1,T2,T3T_1,T_2,T_3 be vectors tangent to Γ10,Γ20,Γ12\Gamma_{10},\Gamma_{20},\Gamma_{12}, respectively, directed away from the junction, normalized so that

∣T1∣=2,∣T2∣=∣T3∣=1.|T_1|=\sqrt{2},\qquad|T_2|=|T_3|=1.

The necessary condition for optimality yields

T1+T2+T3=0.(22)T_1 + T_2 + T_3 = 0. \tag*{(22)}

Taking the scalar product of (4.8) with TiT_i, i,1,2,3i,1,2,3, one obtains

⟨T2,T3⟩=0,⟨T1,T2⟩=⟨T1,T3⟩=−1.\langle T_2,T_3\rangle= 0,\qquad\langle T_1,T_2\rangle= \langle T_1,T_3\rangle= -1.

Thus the two arcs Γ12\Gamma_{12} and Γ20\Gamma_{20} having weight 1 meet orthogonally, while the arc Γ10\Gamma_{10} with weight 2\sqrt{2} forms an angle 3π/43\pi/4 with Γ21\Gamma_{21} and Γ20\Gamma_{20}.

Step 1. Place the two junctions at (0,±h)(0,\pm h), with h>0h > 0. Let α\alpha be the signed angle between the downward vertical and the tangent to the internal arc at the upper junction, positive when the arc bends into the second chamber (see Figure 3).

At the upper junction, the tangent relations just established give

T3=(sin⁡α,−cos⁡α),T2=(cos⁡α,sin⁡α),T1=(−sin⁡α−cos⁡α,cos⁡α−sin⁡α).(23)\begin{aligned} T_3 &= (\sin\alpha,-\cos\alpha),\\ T_2 &= (\cos\alpha,\sin\alpha),\\ T_1 &= (-\sin\alpha-\cos\alpha,\cos\alpha-\sin\alpha). \tag*{(23)} \end{aligned}

Here the direction of T2T_2 is chosen along the exterior arc of the second chamber, and T1T_1 then follows from (4.8). Since both junctions lie on the vertical axis, the exterior circles of Ω1\Omega_1 and Ω2\Omega_2 have radii

2hsin⁡α+cos⁡α,hcos⁡α,(24)\frac{\sqrt{2}h}{\sin\alpha+\cos\alpha},\qquad\frac{h}{\cos\alpha}, \tag*{(24)}

respectively. Imposing that these radii are positive, we obtain the parameter range

−π/4<α<π/2.-\pi/4 < \alpha< \pi/2.

In this step, we prove that the area fraction

q=∣Ω1∣∣Ω1∣+∣Ω2∣q = \frac{|\Omega_1|}{|\Omega_1|+|\Omega_2|}

is a smooth, strictly decreasing function of α\alpha on this interval, with limits 1 and 0 at its two endpoints. This is a quite intuitive fact: when α\alpha increases, from Figure 3 we expect the area fraction of the second chamber to increase. This can be justified by elementary calculations, or the following scaling/uniqueness argument. By contradiction, assume that q(α)q(\alpha) is not a monotone function of α\alpha. Then for two values α1,α2\alpha_1,\alpha_2, one has q(α1)=q(α2)q(\alpha_1)=q(\alpha_2). However, we know that (see e.g. [16]) that for a given qq there exists a unique double bubble satisfying the tangent relations (22) with given total area. Since the angles αi\alpha_i are not changed by dilations, this would contradict the above uniqueness of double bubbles.

Thus the map α↦q(α)\alpha\mapsto q(\alpha) is a smooth bijection from ]−π/4,π/2[]-\pi/4,\pi/2[ onto ]0,1[]0,1[ with nonvanishing derivative. The inverse function theorem gives a smooth inverse q↦α(q)q \mapsto\alpha(q), satisfying

α(0+)=π2,α(1−)=−π4.\alpha(0+) = \frac{\pi}{2}, \qquad\alpha(1-) = -\frac{\pi}{4}.

Step 2. We now prove that, for fixed vv, the function m↦G(m,v)m \mapsto G(m,v) first increases and then decreases. This will imply

G(m,v)≥min⁡{G(m0,v),G(v,v)},m0≤m≤v.G(m,v) \ge\min\{G(m_0,v),G(v,v)\}, \qquad m_0 \le m \le v.

We work with bubbles of prescribed total area vv, translated so that the junctions occur at the points (0,±h)(0,\pm h). Their scale hh is allowed to vary with mm.

We calculate the derivatives of the function D(m,n)\mathcal{D}(m,n), which can be done via a standard Lagrange-multiplier argument. In fact, by the Lagrange multiplier theorem, there exist p1,p2p_1,p_2 such that, calling

J(Ω1,Ω2)≔H1(Γ12)+H1(Γ20)+2⋅H1(Γ10)\mathcal{J}(\Omega_1,\Omega_2) \coloneqq\mathcal{H}^1(\Gamma_{12})+\mathcal{H}^1(\Gamma_{20})+\sqrt{2}\cdot\mathcal{H}^1(\Gamma_{10})

there holds, for an optimal configuration Ω1,Ω2\Omega_1,\Omega_2 with areas m,nm,n, that

δJ(Ω1,Ω2)=p1δ∣Ω1∣+p2δ∣Ω2∣.\delta\mathcal{J}(\Omega_1,\Omega_2)=p_1\delta|\Omega_1|+p_2\delta|\Omega_2|.

Here pip_i must be the curvature of the boundary of the exterior arc of Ωi\Omega_i multiplied by the weight at that point. By (24), the curvatures of the two exterior arcs, multiplied by their respective weights, are

cos⁡α+sin⁡αh,cos⁡αh.\frac{\cos\alpha+\sin\alpha}{h}, \qquad\frac{\cos\alpha}{h}.

Notice in passing that the curvature coefficient on the internal arc, for a displacement from Ω1\Omega_1 towards Ω2\Omega_2, is

sin⁡αh.\frac{\sin\alpha}{h}.

In particular, this is the difference of the two weighted exterior curvatures. Thus

dD=cos⁡α+sin⁡αh d∣Ω1∣+cos⁡αh d∣Ω2∣.(25)d\mathcal{D}=\frac{\cos\alpha+\sin\alpha}{h}\,d|\Omega_1|+\frac{\cos\alpha}{h}\,d|\Omega_2|. \tag*{(25)}

Since ∣Ω1∣=m|\Omega_1|=m and ∣Ω2∣=v−m|\Omega_2|=v-m, this becomes

dG=sin⁡αh dm+cos⁡αh dv.dG=\frac{\sin\alpha}{h}\,dm+\frac{\cos\alpha}{h}\,dv.

Now Step 1 provides a smooth family of minimizing configurations for 0<m<v0<m<v, parametrized by α(q)=α(m/n)\alpha(q)=\alpha(m/n). Keeping vv fixed, we thus obtain

∂G∂m(m,v)=sin⁡α(m/v)h.(26)\frac{\partial G}{\partial m}(m,v)=\frac{\sin\alpha(m/v)}{h}. \tag*{(26)}

Geometrically, transferring area from the second chamber to the first increases the cost when the internal arc bends into the second chamber. The derivative vanishes when the interface is straight, and becomes negative when it bends into the first chamber. Its sign therefore depends only on α\alpha.

By Step 1, α(m/v)\alpha(m/v) decreases strictly from π/2\pi/2 to −π/4-\pi/4 as mm increases from 0 to vv. Since h>0h>0, the right hand side of (26) is positive until the unique configuration with α=0\alpha=0, and becomes negative afterwards. Hence m↦G(m,v)m\mapsto G(m,v) has exactly one interior maximum and no interior minimum.

The function G(m,v)G(m,v) can be easily calculated at the endpoints; in fact, when the total area is equal to vv, the optimal configuration G(0,v)G(0,v) is just a circle, with perimeter weighted by 1, while G(v,v)G(v,v) is again a circle, but with perimeter weighted by 2\sqrt{2}. This yields immediately

G(0,v)=2πv,G(v,v)=22πv.G(0,v)=2\sqrt{\pi}v,\qquad G(v,v)=2\sqrt{2\pi}v.

Since m↦G(m,v)m\mapsto G(m,v) has a unique local maximum in [m0,v][m_0,v], we obtain

G(m,v)≥min⁡{G(m0,v),G(v,v)},m0≤m≤v,G(m,v)\ge\min\{G(m_0,v),G(v,v)\},\qquad m_0\le m\le v,

as claimed.

Step 3. In this step we show that

G(m0,v0)<G(v0,v0).G(m_0,v_0)<G(v_0,v_0).

Consider the bubble with α=π/4\alpha=\pi/4 and h=1h=1. By the tangency relations (23), the circles carrying Γ10\Gamma_{10}, Γ20\Gamma_{20} and Γ12\Gamma_{12} have centers (0,0)(0,0), (1,0)(1,0) and (−1,0)(-1,0), and radii 1, 2\sqrt{2} and 2\sqrt{2}, respectively.

It is easy to see that in this case

∣Ω1∣=π−1,|\Omega_1|=\pi-1,
∣Ω2∣=π+2.|\Omega_2|=\pi+2.

The lengths of the two exterior arcs and the internal arc are, respectively,

Γ1,0=π,Γ20=3π2,Γ12=π2.\Gamma_{1,0}=\pi,\qquad\Gamma_{20}=\frac{3\pi}{\sqrt{2}},\qquad\Gamma_{12}=\frac{\pi}{\sqrt{2}}.

Thus the total area is v=2π+1v=2\pi+1 and the weighted perimeter is

D(π−1,π+2)=2π+3π2+π2=32π.D(\pi-1,\pi+2)=\sqrt{2}\pi+\frac{3\pi}{\sqrt{2}}+\frac{\pi}{\sqrt{2}}=3\sqrt{2}\pi.

Dilating this bubble to total area v0v_0, using (18), gives

G(π−12π+1v0,v0)=32πv02π+1.G\left(\frac{\pi-1}{2\pi+1}v_0,v_0\right)=3\sqrt{2}\pi\sqrt{\frac{v_0}{2\pi+1}}.

Since

m0v0≤14<π−12π+1,\frac{m_0}{v_0}\le\frac{1}{4}<\frac{\pi-1}{2\pi+1},

by Step 1 we obtain that that α≥π/4\alpha\ge\pi/4 whenever

0<m≤(π−1)v0(2π+1)=:m⋆.0 < m \le\frac{(\pi- 1)v_0}{(2\pi+ 1)} =: m^\star.

By Step 2 and continuity at m=0m = 0, the map m↦G(m,v0)m \mapsto G(m,v_0) is increasing on [0,m⋆][0,m^\star]. Therefore we finally get

G(m0,v0)≤G(π−12π+1v0,v0)=32 πv02π+1<22πv0=G(v0,v0).G(m_0,v_0) \le G\left(\frac{\pi- 1}{2\pi+ 1}v_0,v_0\right) = 3\sqrt{2}\,\pi\sqrt{\frac{v_0}{2\pi+ 1}} < 2\sqrt{2\pi v_0} = G(v_0,v_0).

Step 4. We now compare both endpoints from Step 2 with G(m0,v0)G(m_0,v_0). For m0>0m_0 > 0, the first-variation formula in Step 2 yields

∂G∂v(m0,v)=cos⁡α(m0/v)h>0for all v≥v0,\frac{\partial G}{\partial v}(m_0,v) = \frac{\cos\alpha(m_0/v)}{h} > 0 \qquad\text{for all } v \ge v_0,

because −π/4<α<π/2-\pi/4 < \alpha< \pi/2 and h>0h > 0. Thus adding area to the second chamber while keeping the first one fixed increases the minimum cost. For m0=0m_0 = 0, the same monotonicity follows directly from G(0,v)=2πvG(0,v) = 2\sqrt{\pi v}. In either case,

G(m0,v)≥G(m0,v0).G(m_0,v) \ge G(m_0,v_0).

Concerning the other endpoint, Step 3 and v≥v0v \ge v_0 give

G(v,v)=22πv≥22πv0>G(m0,v0).G(v,v) = 2\sqrt{2\pi v} \ge2\sqrt{2\pi v_0} > G(m_0,v_0).

Combining these inequalities with the endpoint-minimum property established in Step 2, we obtain

G(m,v)≥min⁡{G(m0,v),G(v,v)}≥G(m0,v0),G(m,v) \ge\min\{G(m_0,v),G(v,v)\} \ge G(m_0,v_0),

which is the desired conclusion.

An improved lower bound

Define

K:=∫01/2D(s2,s+14) ds,K := \int_0^{1/2} \mathcal{D}\left(s^2,s+\frac{1}{4}\right)\,ds,
μ:=∫01/2(12+2 s) ds=2+28,(27)\mu:= \int_0^{1/2}\left(\frac{1}{2}+\sqrt{2}\,s\right)\,ds = \frac{2+\sqrt{2}}{8}, \tag*{(27)}

and set

c∗:=K24,Λ:=K24μ=(2−2)K2.(28)c_\ast:= \frac{K^2}{4}, \qquad\Lambda:= \frac{K^2}{4\mu} = (2-\sqrt{2})K^2. \tag*{(28)}

The two arguments of D\mathcal{D} correspond to the area at the earlier time and the area gained between the two times, since (s+1/2)2−s2=s+1/4(s+1/2)^2-s^2=s+1/4. A numerical evaluation of the integral defining KK gives that the constant Λ\Lambda yields

Λ≈1.62621170228…\Lambda\approx1.62621170228\ldots

Theorem 4.1 Let γ(⋅)\gamma(\cdot) be a λ\lambda-admissible strategy and let A(t)=L2(Rtγ)A(t)=\mathcal{L}^2(R_t^\gamma), with A(0)>0A(0)>0. If λ≤Λ\lambda\le\Lambda, then

A(t)>c∗t2for every t>0.(29)A(t)>c_*t^2 \qquad\text{for every }t>0. \tag*{(29)}

In particular, the fire cannot be blocked.

Proof. Suppose that (29) fails and let

T=inf⁡{t>0:A(t)≤c∗t2}.T=\inf\{t>0:A(t)\le c_*t^2\}.

Since AA is nondecreasing and A(0)>0A(0)>0, we have T>0T>0 and

A(T)=c∗T2,A(t)>c∗t2for 0<t<T.(30)A(T)=c_*T^2,\qquad A(t)>c_*t^2\quad\text{for }0<t<T. \tag*{(30)}

For almost every t∈(0,T/2)t\in(0,T/2), both tt and t+T/2t+T/2 are good times for Proposition 2.1. Since

c∗t2c∗(t+T/2)2≤14,\frac{c_*t^2}{c_*(t+T/2)^2}\le\frac14,

we can apply Lemma 4.1 with m=A(t)m=A(t), v=A(T)v=A(T). Combining the lemma with (20), Proposition 4.1, and the scaling identity (18), we obtain

c∗ TD((tT)2,tT+14)≤D(A(t),A(t+T/2)−A(t))≤F(t)+F(t+T/2)+λ(T2+2t).\begin{aligned} \sqrt{c_*}\,T\mathcal{D}\left(\left(\frac{t}{T}\right)^2,\frac{t}{T}+\frac14\right) \le\mathcal{D}\left(A(t),A(t+T/2)-A(t)\right) \\ &\le F(t)+F(t+T/2)+\lambda\left(\frac{T}{2}+\sqrt{2}t\right). \end{aligned}

Integrating over (0,T/2)(0,T/2) and using (6) to estimate the right hand side gives

c∗ KT2≤∫0TF(t) dt+λμT2≤A(T)−A(0)+λμT2=(c∗+λμ)T2−A(0).(31)\begin{aligned} \sqrt{c_*}\,KT^2\le\int_0^T F(t)\,\mathrm{d}t+\lambda\mu T^2 \\ &\le A(T)-A(0)+\lambda\mu T^2 \\ &=(c_*+\lambda\mu)T^2-A(0). \tag*{(31)} \end{aligned}

But c∗K=2c∗\sqrt{c_*}K=2c_* and λμ≤Λμ=c∗\lambda\mu\le\Lambda\mu=c_*, hence (31) implies

2c∗T2≤2c∗T2−A(0),2c_*T^2\le2c_*T^2-A(0),

contradicting A(0)>0A(0)>0. □

The choice of c∗c_* is optimal within this quadratic comparison argument. Indeed, with a general coefficient c>0c>0, the same proof gives a contradiction whenever

λ≤cK−cμ=Λ−(c−K/2)2μ.\lambda\le\frac{\sqrt{c}K-c}{\mu} =\Lambda-\frac{(\sqrt{c}-K/2)^2}{\mu}.

The largest threshold is therefore Λ\Lambda, attained at c=c∗c=c_*.

Barriers which are Jordan curves

In this final section we examine the case where the entire barrier Γ=⋃t>0γ(t)\Gamma= \bigcup_{t>0}\gamma(t) is a Jordan curve. When λ>2\lambda> 2 it is well known [7] that there exists an admissible strategy which constructs two arcs of logarithmic spirals and completely surrounds the fire. Once completed, the entire barrier is a Jordan curve.

On the other hand, the following intuitive argument (already sketched in [4]) shows that a strategy which block the fire by constructing a Jordan curve can be admissible only if the construction speed is λ>2\lambda> 2.

Assume that the fire starts from the unit ball centered at the origin. As shown in Figure 4, let xˉ∈Γ\bar{x} \in\Gamma be the location where the last brick of the wall is put in place, say at time TT. Then the entire wall must be constructed before the fire can reach the point xˉ\bar{x}. Since λ\lambda is the construction rate, this implies

Figure 4. If the entire barrier Γ\Gamma is a simple closed curve, then Γ=Γ1∪Γ2\Gamma= \Gamma_{1} \cup\Gamma_{2}, where each of the two portions has length greater than the length of γ0\gamma_{0}. Therefore, the time TT needed for the fire to reach any point of Γ\Gamma is strictly smaller than half of the length of Γ\Gamma.

λT≥H1(Γ).\lambda T \ge\mathcal{H}^{1}(\Gamma).

Let γ\gamma be a path with minimum length, contained inside the region bounded by Γ\Gamma, joining the origin with xˉ\bar{x}. By minimality, the first portion of γ\gamma must be a segment. We prolong this segment backward until it crosses a first point x0∈Γx_{0} \in\Gamma. Call γ0\gamma_{0} this longer path connecting x0x_{0} with xˉ\bar{x}. Moreover, call Γ1,Γ2\Gamma_{1},\Gamma_{2} respectively the two portions of Γ\Gamma, covered when traveling from x0x_{0} to xˉ\bar{x} clockwise and counterclockwise, respectively. Since Γ\Gamma is a simple closed curve, the lengths of these paths satisfy

H1(γ)+1≤H1(γ0)≤min⁡{H1(Γ1),H1(Γ2)}.\mathcal{H}^{1}(\gamma) + 1 \le\mathcal{H}^{1}(\gamma_{0}) \le\min\{\mathcal{H}^{1}(\Gamma_{1}),\mathcal{H}^{1}(\Gamma_{2})\}.

Since T≤H1(γ)T \le\mathcal{H}^{1}(\gamma) (otherwise the fire escapes from xˉ\bar{x} before the wall is completed), we thus conclude

λT=H1(Γ)≥2(H1(γ)+1)≥2(T+1).\lambda T = \mathcal{H}^{1}(\Gamma) \ge2(\mathcal{H}^{1}(\gamma) + 1) \ge2(T + 1).

Hence λ>2\lambda> 2.

We observe, however, that no contradiction is reached if λ>1\lambda> 1 and the barrier Γ\Gamma is not a simple closed curve.

Toward a more rigorous proof, we recall some results from [10]. Let Γ⊂R2\Gamma\subset\mathbb{R}^{2} be a complete rectifiable set. The set reached by the fire at time tt in the presence of the barrier Γ\Gamma is then defined as

RtΓ≔{x(t): x(⋅) is absolutely continuous, x(0)∈R0, ∣x˙(τ)∣≤1 for a.e. τ∈[0,t], x(τ)∉Γ for all τ∈[0,t]}.(32)R_t^\Gamma\coloneqq\left\{x(t):\ x(\cdot)\text{ is absolutely continuous},\ x(0)\in R_0,\ |\dot{x}(\tau)|\le1\text{ for a.e. }\tau\in[0,t],\ x(\tau)\notin\Gamma\text{ for all }\tau\in[0,t]\right\}. \tag*{(32)}

We also define R∞Γ≔⋃t≥0RtΓR_\infty^\Gamma\coloneqq\bigcup_{t\ge0}R_t^\Gamma. We say that the complete, rectifiable set Γ\Gamma is λ\lambda-admissible if, for every t≥0t\ge0, the set

γ(t)≔Γ∩RΓ(t)‾(33)\gamma(t)\coloneqq\Gamma\cap\overline{R^\Gamma(t)} \tag*{(33)}

satisfies (1.1), i.e. it can be constructed within time tt. We now consider:

(BP′) Find an admissible rectifiable set Γ⊂R2\Gamma\subset\mathbb{R}^2 such that the corresponding reachable set R∞ΓR_\infty^\Gamma is bounded.

The main result in [10] shows that, under the assumptions (A1)–(A3), the two formulations of the blocking problem (BP) and (BP′) are equivalent.

Given a rectifiable set Γ\Gamma, we define the corresponding multifunction t↦γ(t)t\mapsto\gamma(t) according to (5.2). Conversely, given an admissible strategy γ(⋅)\gamma(\cdot), we consider the set

Γ≔completion of (γ∞∖R∞γ),γ∞≔⋃t>0γ(t),R∞γ≔⋃t>0Rtγ.(34)\Gamma\coloneqq\text{completion of }(\gamma_\infty\setminus R_\infty^\gamma),\qquad\gamma_\infty\coloneqq\bigcup_{t>0}\gamma(t),\qquad R_\infty^\gamma\coloneqq\bigcup_{t>0}R_t^\gamma. \tag*{(34)}

The following theorem was proved in [10].

Theorem 5.1 In the above setting, there exists an admissible strategy t↦γ(t)t\mapsto\gamma(t) with H1(γ∞)<∞\mathcal{H}^1(\gamma_\infty)<\infty which solves the blocking problem (BP) if and only if there exists an admissible rectifiable set Γ\Gamma which solves (BP′).

We can now prove:

Theorem 5.2 Let Γ\Gamma be a Jordan curve with finite length, blocking a fire initially starting on a bounded open set R0R_0. Then Γ\Gamma can be λ\lambda-admissible only if λ>2\lambda>2.

Proof. 1. Let ℓ=H1(Γ)\ell=\mathcal{H}^1(\Gamma). We will show that every point x∈Γx\in\Gamma can be reached by the fire within time T≔(ℓ−r)/2T\coloneqq(\ell-r)/2. This implies that RΓ(T)⊃ΓR^\Gamma(T)\supset\Gamma. Hence

H1(RΓ(T)‾∩Γ)=H1(Γ)=ℓ>2T.\mathcal{H}^1\left(\overline{R^\Gamma(T)}\cap\Gamma\right)=\mathcal{H}^1(\Gamma)=\ell>2T.

As a consequence, if λ≤2\lambda\le2, then Γ\Gamma is not λ\lambda-admissible.

2. Denote by EE the bounded, open, connected component of R2∖Γ\mathbb{R}^2\setminus\Gamma. Note that its boundary is ∂E=Γ\partial E=\Gamma, while its closure is E‾=E∪Γ\overline{E}=E\cup\Gamma. By a translation, we can assume that the initial burned set R0⊂ER_0\subset E contains the open ball B(0,r)B(0,r) centered at the origin with radius rr. We consider the path distance between points in E‾\overline{E},

dE(x1,x2)≔min⁡{∫01∣x˙(t)∣ dt; x:[0,1]↦E‾ is an absolutely continuous path with x(0)=x1, x(1)=x2}.d_E(x_1,x_2)\coloneqq\min\left\{\int_0^1|\dot{x}(t)|\,dt;\ x:[0,1]\mapsto\overline{E}\text{ is an absolutely continuous path with }x(0)=x_1,\ x(1)=x_2\right\}.

Since Γ\Gamma has bounded length, the existence of a minimizing path follows from a standard compactness argument.

As proved in Lemma 2 of [11], since the Jordan domain EE is simply connected, the maximum distance between two points in E‾\overline{E} is achieved on the boundary:

sup⁡x1,x2∈E‾dE(x1,x2)=sup⁡x1,x2∈ΓdE(x1,x2).(35)\sup_{x_1,x_2\in\overline{E}} d_E(x_1,x_2)=\sup_{x_1,x_2\in\Gamma}d_E(x_1,x_2). \tag*{(35)}
  1. Given any two distinct points x1,x2∈Γx_1,x_2\in\Gamma, let Γ1,Γ2⊂Γ\Gamma_1,\Gamma_2\subset\Gamma be the arcs travelled by moving along Γ\Gamma from x1x_1 to x2x_2 in the clockwise and in the counterclockwise direction, respectively. Clearly ℓ=H1(Γ)=H1(Γ1)+H1(Γ2)\ell=\mathcal{H}^1(\Gamma)=\mathcal{H}^1(\Gamma_1)+\mathcal{H}^1(\Gamma_2), and hence

dE(x1,x2)≤min⁡{H1(Γ1)+H1(Γ2)}≤ℓ2.(36)d_E(x_1,x_2)\leq\min\left\{\mathcal{H}^1(\Gamma_1)+\mathcal{H}^1(\Gamma_2)\right\}\leq\frac{\ell}{2}. \tag*{(36)}
  1. Next, consider any point xˉ∈Γ\bar{x}\in\Gamma. By (36) there is a path γxˉ:[0,1]↦E‾\gamma^{\bar{x}}:[0,1]\mapsto\overline{E} starting from the origin and reaching xˉ\bar{x}, with length ≤ℓ/2\leq\ell/2. Since R0R_0 contains a ball of radius rr centered at 00, the path γxˉ\gamma^{\bar{x}} contains an initial arc connecting the origin to a point x∗∈R0x^*\in R_0 with ∣x∗∣=r/2\lvert x^*\rvert=r/2. Removing this arc, we obtain a shorter path γ∗\gamma^* connecting x∗x^* with xˉ\bar{x}, with length ≤(ℓ−r)/2\leq(\ell-r)/2. Hence

dE(x∗,xˉ)≤ℓ−r2.d_E(x^*,\bar{x})\leq\frac{\ell-r}{2}.
  1. In this final step we prove that, for any ε>0\varepsilon>0, there exists a path γε\gamma_\varepsilon starting inside R0R_0 and reaching a point xˉε∈B(xˉ,ε)\bar{x}_\varepsilon\in B(\bar{x},\varepsilon) within time ≤(ℓ−r)/2\leq(\ell-r)/2, without ever touching the barrier Γ\Gamma.

For sufficiently small δ>0\delta>0, call B(Γ,δ)B(\Gamma,\delta) the open neighborhood with radius δ>0\delta>0 around the set Γ\Gamma and let EδE_\delta be the connected component of R2∖B(Γ,δ)\mathbb{R}^2\setminus B(\Gamma,\delta) which intersects R0R_0. Note that EδE_\delta is simply connected. By choosing δ>0\delta>0 small enough, we also have

B(x∗,r/2)⊂Eδ,Eδ∩B(xˉ,ε)≠∅.(37)B(x^*,r/2)\subset E_\delta,\qquad E_\delta\cap B(\bar{x},\varepsilon)\ne\varnothing. \tag*{(37)}

We now repeat the previous argument, replacing Γ\Gamma with the boundary ∂Eδ\partial E_\delta and xˉ\bar{x} with a point xˉε∈∂Eδ∩B(xˉ,ε)\bar{x}_\varepsilon\in\partial E_\delta\cap B(\bar{x},\varepsilon). This yields the existence of a path γε\gamma_\varepsilon, entirely contained inside EδE_\delta, joining the point x∗∈R0x^*\in R_0 with xˉε\bar{x}_\varepsilon. Moreover, the length of this path satisfies

H1(γε)+r2≤12H1(∂Eδ).\mathcal{H}^1(\gamma_\varepsilon)+\frac{r}{2}\leq\frac{1}{2}\mathcal{H}^1(\partial E_\delta).

Since Γ\Gamma is a Jordan curve with finite length, we have [2]

lim inf⁡δ→0+H1(∂Eδ)=H1(Γ).\liminf_{\delta\to0^+}\mathcal{H}^1(\partial E_\delta)=\mathcal{H}^1(\Gamma).

We thus conclude that

Γ⊂R‾Tfor T=12(H1(Γ)−r2).\Gamma\subset\overline{R}_T\qquad\text{for }T=\frac{1}{2}\left(\mathcal{H}^1(\Gamma)-\frac{r}{2}\right).

Hence Γ\Gamma is not λ\lambda-admissible for any λ≤2\lambda\leq2.

We observe that the condition that Γ=⋃t>0γ(t)\Gamma= \bigcup_{t>0}\gamma(t) is a Jordan curve implies

γ(t)∩Rtγ‾⊆∂u(R2∖Rtγ‾)for all t>0,(38)\gamma(t) \cap\overline{R_t^\gamma} \subseteq\partial^u\left(\mathbb{R}^2 \setminus\overline{R_t^\gamma}\right) \qquad\text{for all } t>0, \tag*{(38)}

where ∂uΩ\partial^u\Omega denotes the boundary of the unbounded component of a set Ω\Omega.

When the construction speed is λ<2\lambda< 2, the next theorem gives a lower bound on the area of the burned set, assuming that (5.7) holds. As before, by ∣Ω∣|\Omega| we denote the Lebesgue measure of a set Ω⊂R2\Omega\subset\mathbb{R}^2.

Theorem 5.3 For any λ<2\lambda< 2, setting cλ:=(2−λ)/4c_\lambda:= (2-\lambda)/4, the following holds. If γ(⋅)\gamma(\cdot) is a λ\lambda-admissible strategy with the property (5.7), then the set burned by the fire satisfies

∣Rtγ∣≥cλt2for all t≥0.(39)|R_t^\gamma| \ge c_\lambda t^2 \qquad\text{for all } t \ge0. \tag*{(39)}

Proof. 1. By contradiction, assume that there exists a τ>0\tau> 0 such that

∣Rτγ∣≤cλτ2,|R_\tau^\gamma| \le c_\lambda\tau^2,

where γ\gamma is a λ\lambda-admissible strategy with the property (5.7). We claim that in this case we can select an earlier time τ′∈[τ/2,τ]\tau' \in[\tau/2,\tau] such that

H1(Fτ′)≤2cλτ.\mathcal{H}^1(F_{\tau'}) \le2c_\lambda\tau.

Otherwise, if H1(Ft)>2cλτ\mathcal{H}^1(F_t) > 2c_\lambda\tau for all t∈[τ/2,τ]t \in[\tau/2,\tau], then by (2.3)

∣Rτγ∣≥∣R0∣+∫τ/2τH1(Ft) dt>∣R0∣+cλτ2,|R_\tau^\gamma| \ge|R_0| + \int_{\tau/2}^{\tau} \mathcal{H}^1(F_t)\,\mathrm{d}t > |R_0| + c_\lambda\tau^2,

reaching a contradiction.

  1. Since τ′≥τ/2\tau' \ge\tau/2, we deduce

H1(Fτ′)≤2cλτ≤4cλτ′.\mathcal{H}^1(F_{\tau'}) \le2c_\lambda\tau\le4c_\lambda\tau'.

Now take

λ′:=λ+4cλ.\lambda' := \lambda+ 4c_\lambda.

and set

Γ:=∂u(R2∖Rτ′γ).\Gamma:= \partial^u\left(\mathbb{R}^2 \setminus R_{\tau'}^\gamma\right).

Since Rτ′γR_{\tau'}^\gamma is connected, Γ\Gamma is a Jordan curve with length

H1(Γ)≤H1(∂Rτ′γ)≤λτ′+4cλτ′=λ′τ′.(40)\mathcal{H}^1(\Gamma) \le\mathcal{H}^1(\partial R_{\tau'}^\gamma) \le\lambda\tau' + 4c_\lambda\tau' = \lambda'\tau'. \tag*{(40)}

We claim that the strategy

γ~(t):={γ(t)∩Rtγ‾if 0≤t<τ′,Γif t≥τ′,(41)\widetilde{\gamma}(t) := \begin{cases} \gamma(t) \cap\overline{R_t^\gamma} & \text{if } 0 \le t < \tau',\\ \Gamma& \text{if } t \ge\tau', \end{cases} \tag*{(41)}

is λ′\lambda'-admissible. Indeed, for 0<s<t<τ′0 < s < t < \tau' we have

Rtγ~=Rtγ,γ~(s)⊆γ~(t).R_t^{\widetilde{\gamma}} = R_t^\gamma,\qquad\widetilde{\gamma}(s) \subseteq\widetilde{\gamma}(t).

Moreover, when s<τ′s < \tau' by (38) one has

γ~(s)=γ(s)∩Rsγ‾⊆γ(τ′)∩Rτ′γ‾⊆∂u(R2∖Rτ′γ‾)=Γ.\widetilde{\gamma}(s) = \gamma(s) \cap\overline{R_s^\gamma} \subseteq\gamma(\tau') \cap\overline{R_{\tau'}^\gamma} \subseteq\partial^u\left(\mathbb{R}^2 \setminus\overline{R_{\tau'}^\gamma}\right) = \Gamma.

This shows that the sets γ~(t)\widetilde{\gamma}(t) satisfy (A1). To prove (A2) we observe that

H1(γ~(t))≤H1(γ(t))=λt<λ′t.\mathcal{H}^1(\widetilde{\gamma}(t)) \leq\mathcal{H}^1(\gamma(t)) = \lambda t < \lambda' t.

while for t≥τ′t \geq\tau' this admissibility condition follows from (40).

3. Adopting the λ′\lambda'-admissible strategy (41), the fire propagation is completely stopped at time τ′\tau'. However, by Theorem 5.2 this is possible only if λ′=λ+4cλ>2\lambda' = \lambda+ 4c_\lambda> 2. Taking cλ=(2−λ)/4c_\lambda= (2-\lambda)/4 we thus obtain a contradiction. This proves the result.

Use of AI tools. The authors acknowledge the use of ChatGPT as an auxiliary research tool. While the idea of considering two distinct times to control the evolution of the burned region arises naturally, the specific idea of using a weighted double-bubble configuration as a comparison was suggested by GPT-6 Astra, together with the weaker estimate of Section 3. The manuscript was written entirely by the authors, who take full responsibility for its mathematical content and correctness.

Acknowledgements. The second author is partially supported by the GNAMPA–INDAM Project 2026, ’Problemi di controllo e regolarità per evoluzione di insiemi’.

References

  1. [1]M. Alfaro, J. Brock, J. Foisy, N. Hodges, and J. Zimba, The standard double soap bubble in ℝ² uniquely minimizes perimeter. Pacific J. Math. 159 (1993), 47–59.
  2. [2]L. Ambrosio, A. Colesanti and E. Villa, Outer Minkowski content for some classes of closed sets. Mathematische Annalen 342 (2008), 727–748.DOI
  3. [3]L. Ambrosio, N. Fusco, and D. Pallara, Functions of Bounded Variation and Free Discontinuity Problems. Oxford University Press, 2000.
  4. [4]A. Bressan, Price offered for a dynamic blocking problem, available at https://bpb-us-e1.wpmucdn.com/sites.psu.edu/dist/c/168431/files/2023/07/prize2.pdf.
  5. [5]A. Barghi and P. Winkler, Firefighting on a random geometric graph, Random Struct. Algor. 46 (2015) 466–477.
  6. [6]S. Bianchini and M. Zizza, Existence of spiral strategies for blocking fire spreading. arXiv:2508.05324arxiv.org/abs/2508.05324
  7. [7]A. Bressan, Differential inclusions and the control of forest fires, J. Differential Equations (special volume in honor of A. Cellina and J. Yorke), 243 (2007), 179-207.
  8. [8]A. Bressan, M. Burago, A. Friend, and J. Jou, Blocking strategies for a fire control problem, Analysis and Applications 6 (2008), 229–246.DOI
  9. [9]A. Bressan and C. De Lellis, Existence of optimal strategies for a fire confinement problem, Comm. Pure Appl. Math. 62 (2009), 789–830.DOI
  10. [10]A. Bressan and T. Wang, Equivalent formulation and numerical analysis of a fire confinement problem, ESAIM; Control, Optim. Calc. Var. 16 (2010), 974–1001.DOI
  11. [11]A. Bressan and T. Wang, The minimum speed for a blocking problem on the half plane, J. Math. Anal. Appl., 356 (2009), 133–144.DOI
  12. [12]A. Bressan and T. Wang, Global optimality conditions for a dynamic blocking problem, ESAIM; Control, Optim. Calc. Var. 18 (2012), 124–156.
  13. [13]C. De Lellis and R. Robyr, Hamilton-Jacobi equations with obstacles, Arch. Ration. Mech. Anal. 200 (2011), 1051–1073.DOI
  14. [14]H. Federer, Geometric Measure Theory. Springer, Berlin, 1969.
  15. [15]R. Klein, E. Langetepe and C. Levcopoulos, A fire-fighter’s problem. In: Proceedings of the 31st International Symposium on Computational Geometry (SoCG 2015).
  16. [16]G. R. Lawlor, Double bubbles for immiscible fluids in ℝⁿ, J. Geom. Anal. 24 (2014), 190–204.
  17. [17]F. Maggi, Sets of Finite Perimeter and Geometric Variational Problems. Cambridge University Press, 2012.DOI
  18. [18]B. Reichardt, Proof of the double bubble conjecture in ℝⁿ. J. Geom. Anal. 18 (2008), 172–191.
  19. [19]J. Schmidt, A lower bound on the blob, preprint 2026, arXiv:2609.13313v1.arxiv.org/abs/2609.13313
  20. [20]T. Wang, Optimality conditions for a blocking strategy involving delaying arcs, J. Optim. Theory Appl. 152 (2012), 307–333.DOI

Paper details

vlnc:W_wdnvteqn
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 6

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents