Abstract
Globally optimal free flight trajectory optimization can be achieved with a combination of discrete and continuous optimization. A key requirement is that Newton’s method for continuous optimization converges in a sufficiently large neighborhood around a minimizer. We show in this paper that, under certain assumptions, this is the case.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Around the world countries are implementing Free Flight airspaces that allow aircraft to choose their own route, as opposed to being restricted to a predetermined three-dimensional network. The primary factors that influence costs are time and fuel consumption, which are closely interrelated [1]. Based on the relative weights of these factors (cf. cost index) the optimal airspeed can be determined, which typically remains largely constant [2,3,4]. Additionally, the vertical flight path can usually be predetermined using aircraft performance data [5]. Consequently, the problem can be well approximated in a way proposed by Zermelo in 1931 [6], which involves finding the most efficient trajectory from point A to B for an aircraft flying at a constant airspeed in a given two-dimensional wind field.
The Free Flight Trajectory Optimization Problem is usually solved using direct or indirect methods from Optimal Control [5, 7,8,9,10,11]. These are highly efficient, but suffer from one key drawback: They only converge locally and are thus dependent on a sufficiently good starting point. This makes such methods, used as a standalone, incapable of meeting airlines’ high expectations regarding the global optimality of routes.
In [12,13,14] a deterministic two-stage algorithm was proposed that combines discrete and continuous optimization in order to find a globally optimal solution to the free flight trajectory optimization problem. With this approach the exponential complexity of other branch and bound based algorithms is circumvented.
The primary objective of the first stage is to obtain a finite sample in a systematic manner that adequately covers the search space. This deterministic approach eliminates the potential for infinite runtime, which may occur when using stochastic global optimization algorithms, such as Particle Swarm Optimization, Simulated Annealing, or Monotonic Basin Hopping [15,16,17,18].
One approach is to create a locally dense directed graph with a specific density determined by the node spacing \(h\) and connectivity length \(\ell\), thereby implicitly defining the sample. The instances can then be selected in order of quality by applying Yen’s algorithm [19] to calculate the kth shortest paths.
Promising paths serve as initial guesses for a subsequent refinement stage in which a continuous solution to the problem is calculated up to the desired accuracy.
Analytical evidence and numerical experiments have demonstrated that the new hybrid algorithm has a time complexity of \(\mathcal {O}(\ell ^{-1})\), making it superior to the conventional purely discrete approach, which has a time complexity of \(\mathcal {O}(\ell ^{-6})\) [12]. In this context, \(\ell\) refers to the maximum arc length in a graph and the discretization length in a continuous optimization scenario. Thus, \(\ell ^{-1}\) serves as a comparable metric for the precision of the solution.
The present paper is concerned with the second stage. One way to generate a continuous solution is to apply Newton’s method to the first order necessary conditions (the KKT-conditions) — an approach commonly referred to as Newton-KKT or Sequential Quadratic Programming (SQP) (see, e.g., [20]). It is now shown that there is a quantifiable domain around a global optimum such that Newton-KKT converges if initialized accordingly.
Since the computational effort of the first graph-searching stage depends exclusively on the problem instance, i.e., the wind conditions, the algorithm asymptotically inherits the super fast convergence rates of the Newton-KKT method.
The paper is structured as follows. After defining the problem and introducing a formulation that is convenient for the analytical discussion in Section 2, we formally state the necessary and sufficient conditions as well as the Newton-KKT approach in Section 3. The proof of convergence is provided in Section 4 followed by a conclusion emphasizing the impact on previous and future work.
2 The Free Flight Trajectory Optimization Problem
2.1 Notation
Throughout this article lower case subscripts, e.g., \(x_t\) or \({\xi _\tau }\), denote partial derivatives, while total derivatives are indicated by primes, e.g., \(T'\) or \(f'\). Locally and globally optimal quantities are indicated by single and double superscript stars, respectively, e.g., \(\xi ^\star\) or \(\xi ^{\star \star }\). If not stated otherwise, we assume \(\Vert \cdot \Vert\) to denote the \(l^2\)-norm. Accordingly, we use the following quantitative definition of the \(L^\infty\)-norm in terms of the \(l^2\)-norm.
Definition 1
Let \(f: \Omega \mapsto {\mathbb {R}}^n\). Then we define
2.2 Problem Statement
Neglecting any traffic flow restrictions, we consider Lipschitz-continuous flight paths \(\xi \in C^{0,1}(]0,1[,{\mathbb {R}}^2)\) connecting origin \(\xi (0)=x_O\) and destination \(\xi (1) =x_D\). By Rademacher’s theorem, such paths are almost everywhere differentiable, and moreover contained in the Sobolev space \(W^{1,\infty }(\mathopen ]0,1\mathclose [,{\mathbb {R}}^2)\).
A short calculation reveals that an aircraft travelling along such a path \(\xi\) with constant airspeed \({\overline{v}}\) through a three times continuously differentiable wind field \(w\in C^3({\mathbb {R}}^2,{\mathbb {R}}^2)\) with bounded magnitude \(\Vert w(x)\Vert < {\overline{v}}\) reaches the destination after a flight duration
with \(\xi _\tau\) denoting the time derivative of \(\xi\) and
Among these paths \(\xi\), we need to find one with minimal flight duration \(T(\xi )\), since that is essentially proportional to fuel consumption [1]. This classic of optimal control is known as Zermelo’s navigation problem [6]. It can easily be shown that in case of bounded wind speed, the optimal trajectory cannot be arbitrarily longer than the straight connection of origin and destination. Hence, every global minimizer is contained in an ellipse \(\Omega \subset {\mathbb {R}}^2\) with focal points \(x_O\) and \(x_D\).
The flight duration T as defined in (2) is based on a time reparametrization from actual flight time \(t\in [0,T]\) to pseudo-time \(\tau \in [0,1]\) according to the actual flight trajectory \(x(t) = \xi (\tau (t))\) such that \(\Vert x_t(t)-w(x(t))\Vert = {\overline{v}}\), where \(x_t\) denotes the so-called ground speed, i.e., the derivative of position x with respect to the unscaled time t. As a consequence, the actual parametrization of \(\xi\) in terms of pseudo-time \(\tau\) is irrelevant for the value of T. Calling two paths \(\xi ,{\tilde{\xi }}\) equivalent if there exists a Lipschitz-continuous bijection \(r:\mathopen ]0,1\mathclose [\rightarrow \mathopen ]0,1\mathclose [\) such that \(\xi (r(\tau )) = {\tilde{\xi }}(\tau )\), we can restrict the optimization to equivalence classes. Moreover, every equivalence class contains a representative with constant ground speed \(\Vert {\xi _\tau }(\tau )\Vert =L\) for almost all \(\tau\), that can be obtained from any \({\tilde{\xi }}\) with \(\Vert {\tilde{\xi }_\tau }(\tau )\Vert \ne 0 ~ \forall \tau\) via
Hence, we introduce \(z:=(L,\xi ) \in Z:= {\mathbb {R}}\times X\) and the affine space of valid trajectories
and subsequently consider the equivalent constrained minimization problem
with
If the constraint is satisfied, L also represents the path length, since
Note that \(T:X\rightarrow {\mathbb {R}}\) is Fréchet differentiable with respect to the corresponding linear space
of directions \({\delta \xi }\) with zero boundary values, that consequently do not change origin and destination, equipped with the norm
Further we define the linear space
and equip the spaces Z and \({\delta Z}\) with the norms
3 Continuous Optimization: Newton-KKT
In order to find a continuous solution to the free flight optimization problem (6) we apply Newton’s method to the first order necessary conditions (the KKT-conditions), which is also known as sequential quadratic programming (SQP). Before we formally introduce Newton’s method, we discuss the necessary and sufficient conditions for optimality, which also defines the goal of the presented algorithm.
3.1 Optimality Conditions
3.1.1 Necessary Conditions
The goal of the present paper is to find an isolated globally optimal solution \(\xi ^{\star \star }\) to (6) that satisfies \(T(\xi ^{\star \star }) \le T(\xi ) \; \forall \xi \in X\), contrary to a local optimizer \(\xi ^{\star }\) that is only superior to trajectories in a certain neighborhood, \(T(\xi ^{\star }) \le T(\xi ) \; \forall \xi \in \mathcal {N}(\xi ^{\star }) \subseteq X\). An isolated global minimizer satisfies the necessary Karush-Kuhn-Tucker (KKT) optimality conditions [21] given that it is a regular point, which is always the case, as confirmed by the following Theorem.
Theorem 1
Let \(z=(L,\xi )\in Z\) with \(L>0\) and assume there is a direction \(u\in {\mathbb {R}}^2\) and \(c>0\) such that \({\xi _\tau }^T u \ge c\) almost everywhere. Then, \(h'(z):{\delta Z}\rightarrow L^\infty (]0,1[)\) is surjective, i.e., \(z\) is regular.
Proof
Let \(f\in L^\infty (]0,1[)\) be given and \(b:={\xi _\tau }^T u\ge c\). We set
and
Due to \(b\ge c\) almost everywhere, \(b^{-1}\) is bounded and hence \(g,{\xi _\tau }\in L^\infty (]0,1[)\). By construction, \(\int _0^1 {\delta \xi _\tau }\,d\tau = 0\) holds, such that \({\delta z}= ({\delta L},{\delta \xi })\in {\delta Z}\).
Now we obtain
and thus the claim.
For \(\lambda \in \Lambda ^*\), the Lagrangian is defined as
The KKT-conditions guarantee for a regular minimizer \(z^{\star \star }\) the existence of a Lagrange multiplier \(\lambda ^{\star \star }\in \Lambda ^*\), such that
hold, where \({\delta z}:= ({\delta L},{\delta \xi }) \in {\delta Z}\). In our case, these necessary conditions read
Let us for a moment consider the unconstrained problem analogous to (6),
Any global minimizer \(\tilde{\xi }^{\star \star }\) of (15) is clearly non-isolated due to possible reparametrizations of the time. Let \(\xi ^{\star \star }\) denote the equivalent trajectory with constant ground speed, i.e., \(\Vert {\xi _\tau }^{\star \star }(\tau )\Vert =L^{\star \star }\) for almost all \(\tau\). Both solutions \(\tilde{\xi }^{\star \star },\xi ^{\star \star }\) satisfy the first order necessary condition
Moreover, \(\xi ^{\star \star }\) — together with \(L^{\star \star }\) from (8) — is a global minimizer of the constrained problem, which indicates that the ground speed constraint (7) is only weakly active. We confirm this by showing that the corresponding Lagrange multipliers \(\lambda ^{\star \star }\) vanish.
Lemma 2
Let \(z^{\star \star } = (\xi ^{\star \star },L^{\star \star })\) be a global minimizer of (6). Then, this solution together with
satisfies the necessary conditions (14).
Proof
Since \(\xi ^{\star \star }\) is also a global minimizer of the unconstrained problem, the necessary condition (16) states that \(T'(\xi ^{\star \star }){\delta \xi }=0\). The term \(\int _0^1 \lambda ^{\star \star }\left( {\delta \xi _\tau }^T {\xi _\tau }^{\star \star }- {\delta L}\,L^{\star \star }\right) d\tau\) of (14a) vanishes for \(\lambda ^{\star \star }=0\). (14b) is satisfied because \(\Vert {\xi _\tau }^{\star \star }\Vert = L^{\star \star }\) for almost all \(\tau \in ]0,1[\).
3.1.2 Sufficient Conditions
Now we turn to the second order sufficient conditions for optimality. In general, a stationary point \((z^{\star },\lambda ^{\star })\) is a strict minimizer, if, in addition to the necessary conditions above, the well known Ladyzhenskaya-Babuška-Brezzi (LBB) conditions (e.g., [22]) are satisfied, which comprise (a) the so-called inf-sup condition and (b) the requirement that the Lagrangian’s Hessian regarding \(z\), \(\mathcal {L}_{zz}\), need be positive definite on the kernel of \(h'\).
The inf-sup condition states that for the minimizer \(z^{\star }\) there is a \(\kappa > 0\) such that
Formally, the second part of the LBB-conditions requires that there is a \({\underline{\mathcal {B}}}> 0\) such that
for any \({\delta z}\in {\delta Z}\) that satisfies
In the present case, this reads
for any \({\delta z}\in {\delta Z}\) such that
In case of a global minimizer \(z^{\star \star } = (\xi ^{\star \star }, L^{\star \star })\), this can be reduced using \(\lambda ^{\star \star }=0\) from Lemma 2. Moreover, the constraint is equivalent to requiring that \({\delta \xi _\tau }^T{\xi _\tau }^{\star \star }= {\delta L}\,L^{\star \star }\) almost everywhere. With this, we conclude that for any isolated global minimizer \(z^{\star \star }\) of (6) that satisfies the inf-sup condition, there exists a \({\underline{\mathcal {B}}}>0\) such that
for any \({\delta z}\in {\delta Z}\) such that \({\delta \xi _\tau }^T{\xi _\tau }^{\star \star }= {\delta L}\,L^{\star \star }\) almost everywhere.
It is important to note that the second order sufficient conditions are formulated in a \(L^2\)-setting, while differentiability only holds in \(L^\infty\). This is known as two-norm-discrepancy [23].
3.2 Newton’s Method
In order to provide a more compact notation, we use \(\chi := (z,\lambda ) \in Z\times \Lambda ^* =: Y\) in this context and define \(F\) as the total derivative of the Lagrangian,
On \(Y\) we define the following norms,
The problem is now to find a \(\chi ^{\star \star }\) such that the first order necessary conditions for optimality as stated in (14) are satisfied, which translates to
Applying Newton’s method, we iteratively solve
for \(\Delta \chi ^k\) and proceed with \(\chi ^{k+1} \leftarrow \chi ^k + \Delta \chi ^k\), starting with some initial value \(\chi ^0\). In other words, in every iteration we need to find \(({\Delta z}^k, {\Delta \lambda }^k)\) such that
4 Proof of Convergence
On the way to prove the existence of a non-empty domain \(\mathcal {B}(\chi ^{\star \star },R)\) such that Newton’s method as defined in Section 3.2 converges to the corresponding global minimizer \(\chi ^{\star \star }\), if initialized with a starting point within this neighborhood, we first prove that the KKT-operator \(F'\) is invertible and that the Newton step \(\Delta \chi ^k\) is always well defined. Essentially, this is the case if the LBB-conditions as given in (18) and (20) are satisfied. Hence, we will show that there is a \(R>0\) such that the inf-sup condition is satisfied and that the Lagrangian is positive definite on the kernel of the constraints for any \(\chi \in \mathcal {B}(\chi ^{\star \star },R)\) . Further, we show that an affine covariant Lipschitz condition holds, which finally helps to complete the proof.
Before we get there, we recall the following Lemma from [13, Lemma 7] which provides a bound for the path length of a global minimizer.
Lemma 3
Let \(z^{\star \star } = (L^{\star \star },\xi ^{\star \star })\) be a global minimizer of (6), let \(\Vert w\Vert _{L^{\infty }(\Omega )} \le {\overline{c}}_0\), and define \(\tilde{L} = \Vert x_D-x_O\Vert\). Then it holds that
As most of the subsequent results hold in a \(L^\infty\)-neighborhood of a minimizer, we introduce the following notation.
Definition 1 We call the \(L^\infty\)-neighborhood of a point \(z\in Z\) or \(x\in Y\),
respectively.
Moreover, we provide three simple yet useful bounds that hold in such a \(L^\infty\)-neighborhood of a minimizer.
Lemma 4
Let \(\chi ^{\star \star }= (z^{\star \star }, \lambda ^{\star \star })\) be a global minimizer of (6) and the corresponding Lagrange multipliers. Then for every \(\chi \in \mathcal {B}(\chi ^{\star \star },R)\) it holds that
Proof
The first two inequalities follow immediately, since a global minimizer satisfies the constraint from (6). The latter two are a direct consequence of Lemma 2.
4.1 Inf-Sup Condition
We now show that the inf-sup condition, introduced in (18), holds in a certain neighborhood around a global minimizer. First, however, we point out that deviations \({\delta \xi }\) and \({\delta \xi _\tau }\) from a trajectory are inherently related and that the former is always bounded by the latter.
Theorem 5
(Wirtinger's inequality)
Let \({\delta \xi }\in H^1_0(]0,1[)\). Then
holds.
Theorem 6
Let \(z^{\star \star }\) be a global minimizer of (6). Further, let there be a constant \(c>0\) and some direction \(u\in {\mathbb {R}}^2\) with \(\Vert u\Vert =1\) such that \(u^T {\xi _\tau }^{\star \star }\ge c\) for almost all \(\tau \in ]0,1[\). Then for any \(z=(L,\xi )\in \mathcal {B}(z^{\star \star }, R)\) with \(R < c\) there is some \(\kappa > 0\) such that
with
Proof
For \(f\in L^2(]0,1[)\) we define
respectively, such that \((\overline{f}, \tilde{f})_{L^2(]0,1[)} = 0\) and
With
we choose \({\delta \xi _\tau }= \frac{1}{2}\tilde{\lambda }u\) and \({\delta L}= \frac{1}{2L}\left( \overline{b\tilde{\lambda }}- (c-R) \overline{\lambda }\right)\). Note that \({\delta \xi }\in {\delta X}\) holds. For this choice, we obtain for \({\delta z}=({\delta L},{\delta \xi })\)
Moreover, we have
and, since clearly \(c \le L^{\star \star }\),
which implies
Consequently,
yields the claim.
4.2 Positive Definiteness of the Lagrangian
The next step in order prove invertibility of the KKT-operator \(F'(\chi )\), (24), is to show that the second partial derivative of the Lagrangian \(\mathcal {L}(\chi )\), (13), with respect to the state \(z\) is positive definite on the kernel of the linearized constraints. On the way we derive a similar result for the objective \(T(\xi )\), (2) for which we first derive an upper bound for its third derivative.
Lemma 7
Let \(\Vert w\Vert _{L^{\infty }(\Omega )} \le {\overline{c}}_0 \le {\overline{v}}/\sqrt{5}\), \(\Vert w_x\Vert _{L^{\infty }(\Omega )} \le {\overline{c}}_1\), \(\Vert w_{xx}\Vert _{L^{\infty }(\Omega )} \le {\overline{c}}_2\), and \(\Vert w_{xxx}\Vert _{L^{\infty }(\Omega )} \le {\overline{c}}_3\) and define \({\underline{v}}^2:= {\overline{v}}^2 - {\overline{c}}_0^2\). Then, for any \(\xi \in X\), the third directional derivative of \(f\) as given in (3) is bounded by
with \({\overline{\gamma }}_i\ge 0\), \(i\in 0,\dots ,5\), given as
The proof can again be found in the Appendix. With this result we can derive a bound for the third directional derivative of \(T\).
Theorem 8
Let \((L^{\star \star }, \xi ^{\star \star })\) be a global minimizer of (6) and define \(\tilde{L}:= \Vert x_D-x_O\Vert\) and \({\Delta \xi }:= \xi -\xi ^{\star \star }\). Moreover, let \(\Vert w(p)\Vert \le {\overline{c}}_0 \le {\overline{v}}/ \sqrt{5}\), \(\Vert w_x(p)\Vert \le {\overline{c}}_1\), \(\Vert w_{xx}(p)\Vert \le {\overline{c}}_2\), and \(\Vert w_{xxx}(p)\Vert \le {\overline{c}}_3\) for every \(p\in \Omega\). Then, for any \(\xi \in X\) with \(\Vert {\Delta \xi }\Vert _{X^{\infty }} \le R < \tilde{L}\), it holds that
with \(\Vert\Delta\xi\Vert_{C^{0,1}(]0,1[)} = \Vert {\Delta\xi}\Vert_{L^{\infty}({]0,1[})} + \Vert\Delta\xi_\tau\Vert_{L^\infty(]0,1[)}\) and
and \({\overline{\gamma }}_0,\dots ,{\overline{\gamma }}_5\) as given in Lemma 7 above.
Proof
From the definition of T in (2), we know that
Inserting the bound from Lemmas 4 and 7 above and using Young’s inequality yields
Having bounded the third derivative of T, we can estimate the potential decay of \(T''\) and thus derive a lower bound for the size of this neighborhood. Similarly, we can bound \(h''\) and hence \(\mathcal {L}_{zz}\).
Theorem 9
Let \(\Vert w\Vert _{L^{\infty }(\Omega )} \le {\overline{c}}_0 < {\overline{v}}/\sqrt{5}\), \(\Vert w_x\Vert _{L^{\infty }(\Omega )} \le {\overline{c}}_1\), \(\Vert w_{xx}\Vert _{L^{\infty }(\Omega )} \le {\overline{c}}_2\), and \(\Vert w_{xxx}\Vert _{L^{\infty }(\Omega )} \le {\overline{c}}_3\) and define \(\tilde{L}:= \Vert x_D-x_O\Vert\). Moreover, let \(\chi ^{\star \star }:= (z^{\star \star }, \lambda ^{\star \star })\) be a globally optimal solution to problem (6), that satisfies the necessary and sufficient conditions (14), (18), and (20) with \({\underline{\mathcal {B}}}>0\). Then there is a \(0< R < \min \left\{ \frac{{\underline{\mathcal {B}}}}{2{\overline{\Gamma }}},\; \frac{{\underline{\mathcal {B}}}}{40},\; \frac{\tilde{L}}{2} \right\}\) with \({\overline{\Gamma }}\) from Theorem 8 such that
holds for any \(\chi \in \mathcal {B}(\chi ^{\star \star }, R)\) and any \({\delta z}\in {\delta Z}\) such that \({\xi _\tau }^T {\delta \xi _\tau }= L{\delta L}\) holds almost everywhere.
Proof
Let \({\Delta \xi }:= \xi -\xi ^{\star \star }\) and note that \(\Vert {\Delta \xi }\Vert _{L^{\infty }({]0,1[} )} \le \Vert {\Delta z}\Vert _{Z^{\infty }} \le R < \frac{{\underline{\mathcal {B}}}}{2{\overline{\Gamma }}}\). Then we obtain
Further, we point out that
which together with the bounds from Lemma 4 yields
Together, these bounds yield the claim with
4.3 Upper Bound for the Lagrangian
As a counterpart to the previous Lemma, we also derive an upper bound for \(L_{zz}\) close to a minimizer. Again we start with the underlying function \(f\) in order to bound the error in the objective function \(T\) .
Lemma 10
Let \(\Vert w\Vert _{L^{\infty }(\Omega )} \le {\overline{c}}_0 \le {\overline{v}}/\sqrt{5}\), \(\Vert w_x\Vert _{L^{\infty }(\Omega )} \le {\overline{c}}_1\), and \(\Vert w_{xx}\Vert _{L^{\infty }(\Omega )} \le {\overline{c}}_2\). Moreover, let \({\underline{v}}^2:= {\overline{v}}^2 - {\overline{c}}_0^2\). Then, for any \(\xi \in X\), the second directional derivative of \(f\) as given in (3) is bounded by
with
The proof can be found in the Appendix.
Theorem 11
Let \(z^{\star \star } = (L^{\star \star },\xi ^{\star \star })\) be a global minimizer of (6) and \({\Delta z}:= z - z^{\star \star }\). Moreover, let \(\Vert w\Vert _{L^{\infty }(\Omega )} \le {\overline{c}}_0 \le \bar{v} / \sqrt{5}\), \(\Vert w_x\Vert _{L^{\infty }(\Omega )} \le {\overline{c}}_1\), and \(\Vert w_{xx}\Vert _{L^{\infty }(\Omega )} \le {\overline{c}}_2\). Also define \({v}^2:= {\overline{v}}^2 - {\overline{c}}_0^2\) and \(\tilde{L}:= \Vert x_D-x_O\Vert\). Then, for any \(z\in \mathcal {B}(z^{\star \star }, R)\) , the second directional derivative of T as defined in (2) is bounded by
with \({\overline{\mathcal {B}}}:= {\overline{\beta }}_1 + \max \left\{ \left( \frac{{\overline{v}}+{\overline{c}}_0}{{\overline{v}}-{\overline{c}}_0} \tilde{L} + R\right) {\overline{\beta }}_0, \; \frac{{\overline{\beta }}_2}{\tilde{L} + R} \right\}\) and \({\overline{\beta }}_0,{\overline{\beta }}_1,{\overline{\beta }}_2\) as defined in Lemma 10.
Proof
From the definition of T in (2) we know that
which, together with the bounds from Lemmas 4 and 10 as well as Young’s inequality, then leads to
Theorem 12
Let \(\chi ^{\star \star }= (z^{\star \star }, \lambda ^{\star \star })\) be a global minimizer of (6) and the corresponding Lagrange multipliers. Then for every \(\chi \in \mathcal {B}(\chi ^{\star \star }, R)\) and every \({\delta z}\in {\delta Z}\) it holds that
with \({\overline{\mathcal {B}}}(R)\) from Theorem 11.
Proof
Using the bound from Theorem 11 and Young’s inequality, we get
4.4 Invertibility of the KKT-Operator
Using the previous three results, which together state the existence of a neighborhood around a minimizer such that the LBB-conditions are satisfied, we are now ready to prove that the KKT-operator \(F'\) is invertible.
Lemma 13
Let \(\chi ^{\star \star }= (z^{\star \star }, \lambda ^{\star \star })\) be a global minimizer of (6), that satisfies the first and second order conditions for optimality with some \({\underline{\mathcal {B}}}> 0\), and the corresponding Lagrange multipliers. Further, let there be a u with \(\Vert u\Vert =1\) such that \(u^T {\xi _\tau }^{\star \star }\ge c > 0\) for almost all \(\tau \in ]0,1[\). Then for \(F\) as given in (21) it holds that
for every \(\chi =(z,\lambda )\in \mathcal {B}(\chi ^{\star \star }, R)\) and
and \({\overline{\mathcal {B}}}(R)\) and \(\kappa (R)\) as given in Theorems 6 and 11, respectively.
Proof
The proof builds on some prerequisites that have been established above and are briefly summarized.
-
(i)
In Theorem 6 it was proved that the inf-sup condition is satisfied:
$$\begin{aligned} \underset{{\delta \lambda }\in L^2(]0,1[)}{\inf }\ \; \underset{{\delta z}\in {\delta Z}}{\sup }\ \; \frac{\langle {\delta \lambda }, h'(z)[{\delta z}]\rangle }{ \Vert {\delta z}\Vert _{Z^{2}} \Vert {\delta \lambda }\Vert _{L^{2}({]0,1[} )} } \ge \kappa > 0. \end{aligned}$$ -
(ii)
In Theorem 9 it was proved that \(\mathcal {L}_{zz}\) is positive definite on the kernel of the constraints, i.e.,
$$\begin{aligned} \mathcal {L}_{zz}(\chi )[{\delta z}]^2 = T''(\xi )[{\delta \xi }]^2 + \langle h''(z)[{\delta z}]^2 \rangle \ge \frac{{\underline{\mathcal {B}}}}{4} \Vert {\delta z}\Vert _{Z^{2}} ^2 \end{aligned}$$for all \({\delta z}\in {\delta Z}\) such that \(h'(z)[{\delta z}] = 0\).
-
(iii)
In Theorem 12 it was proved that \(\mathcal {L}_{zz}\) is bounded from above as
$$\begin{aligned} |\mathcal {L}_{zz}(\chi )[{\delta z}]^2| = |T''(\xi )[{\delta \xi }]^2 + \langle h''(z)[{\delta z}]^2 \rangle | \le ({\overline{\mathcal {B}}}+ R) \Vert {\delta z}\Vert _{Z^{2}} ^2 . \end{aligned}$$
Under these conditions, it follows from Brezzi’s Splitting Theorem [22, Thm. 4.3] that \(F'(x)\) is isomorphic. Further, it can be shown that for every right hand side F(x) of the saddle point problem (24) there is exactly one solution \(({\Delta z}, {\Delta \lambda })\) with
With \(\Vert F(\chi )\Vert = \Vert T'(\xi ) + \langle \lambda , h'(z) \rangle \Vert _{Z^{2}} ^2 + \Vert h(z)\Vert _{L^{2}({]0,1[} )} ^2\) follows that
which directly yields
with \(\omega _1 = \sqrt{2} \max \left\{ \frac{4}{{\underline{\mathcal {B}}}}, \frac{1}{\kappa }\left( 1+\frac{4({\overline{\mathcal {B}}}+R)}{{\underline{\mathcal {B}}}}\right) , \frac{{\overline{\mathcal {B}}}+R}{\kappa ^2} \right\}\). This completes the proof, since
4.5 Lipschitz Constant
We are on the verge of presenting a Lipschitz constant for the free flight problem. To accomplish this, we introduce an additional bound in the form of a Lemma. This bound incorporates the constant \(\hat{\mathcal {B}}\), which is derived in the Appendix (Lemma 18). It serves to define an upper limit on the second derivative of \(f\) as defined in 3. Its value is contingent upon the overall characteristics of the wind field.
Lemma 14
Let \(\chi ^{\star \star }= (z^{\star \star }, \lambda ^{\star \star })\) be a global minimizer of (6) and the corresponding Lagrange multipliers. For any \(\chi _{i\in \{1,2\}} \in \mathcal {B}(\chi^{\star\star}, R)\) there is a \(\hat{\mathcal {B}}\) such that
with
Proof
From Lemma 4 it directly follows that
Using these bounds as well as the Cauchy-Schwarz inequality and Young’s inequality, we show that for any \(\delta \chi \in {\delta Z}\times L^2(]0,1[)\) with \(\Vert \delta \chi \Vert _{L^{2}({]0,1[} )} \le 1\) it holds that
as well as
and
As shown in Lemma 18 in the Appendix, there is a \(\hat{B}<\infty\) such that
which provides the following bound, as
Finally, we use the bounds derived above to show that for any \(\delta x\) with \(\Vert \delta x\Vert _{Y^{2}} \le 1\) it holds that
with
This directly yields the claim, as
4.6 Convergence of Newton’s Method
We are now ready to connect the results outlined above to prove that the Newton-KKT method applied to the free flight optimization problem (6) converges to a global minimizer as characterized in Section 3.1 provided that there is a \(u\in {\mathbb {R}}^2\) with \(\Vert u\Vert =1\) such that \(u^T{\xi _\tau }^{\star \star }\ge c\). Roughly speaking, the optimal route needs to head towards the destination, dominating any route that involves flying the opposite direction. It is intuitively clear that this holds even for relatively strong wind conditions.
Theorem 15
Let \(\chi ^{\star \star }= (z^{\star \star }, \lambda ^{\star \star })\) be a global solution of (6) that satisfies the first and second order conditions for optimality with \({\underline{\mathcal {B}}}>0\). Moreover let there be a \(c>0\) and a \(u\in {\mathbb {R}}^2\) with \(\Vert u\Vert =1\) such that \(u^T{\xi _\tau }^{\star \star }\ge c\) for almost all \(\tau \in ]0,1[\). Finally, let \(\omega :=\omega _1\omega _2\), as given in Lemmas 13 and 14.
Then there is a \(R_C>0\), such that the ordinary Newton iterates defined in Section 3.2 converge to \(\chi ^{\star \star }\) at an estimated rate
if initialized with \(\chi ^0\in \mathcal {B}(\chi ^{\star \star }, R_C){}\) and provided that the iterates \(\chi ^k\) remain in \(\mathcal {B}(\chi ^{\star \star }, R_C){}\). Moreover, \(\chi ^{\star \star }\) is unique in \(\mathcal {B}(\chi ^{\star \star }, R_C){}\).
Proof
In Theorems 6, 9 and 12 we showed that the inf-sup condition is satisfied, that, \(\mathcal {L}_{zz}(\chi )\) is positive definite on the kernel of the constraint for all \(x\in \mathcal {B}(\chi ^{\star \star }, R_C){}\), and that it is bounded from above. Consequently, \(F'(\chi )\) is invertible with
as confirmed in Lemma 13. Further, it follows from Lemmas 13 and 14 that
for \(\chi _1,\chi _2 \in \mathcal {B}(\chi ^{\star \star }, R_C){}\). It is clear that since \(\omega _1\) is bounded and \(\omega _2 = (8 + \hat{\mathcal {B}}) R\), there is a \(R_C>0\) such that \(\omega := \omega _1\omega _2 < 2\). We now define \(e_k:= \chi ^k - \chi ^{\star \star }\) and proceed for \(\mu \in ]0,1[\) as follows:
which yields the claim with \(\mu =1\) as
In order to prove uniqueness in \(\mathcal {B}(\chi ^{\star \star }, R_C){}\), assume there is a second solution \(\chi ^{\star }\ne \chi ^{\star \star }\) with \(F(\chi ^{\star })=0\) and \(\chi ^{\star }\in \mathcal {B}(\chi ^{\star \star }, R_C){}\). Initialized with \(\chi ^0:= \chi ^{\star }\) it certainly holds that \(\chi ^1 = \chi ^{\star }\). However, from (47) we obtain
due to \(\omega <2\), which yields a contradiction.
5 Conclusion
It has been demonstrated that the Newton-KKT method can be used to solve the free flight trajectory optimization problem under certain conditions. These conditions are i) the requirement for the iterates to remain within a \(L^\infty\)-neighborhood of the solution, and ii) a starting point that is sufficiently close to the solution. Such a suitable starting point can be found efficiently by calculating shortest paths on a specific graph [13]. Hence an important tool for efficient deterministic global optimization of the free flight problem has been established.
Data Availability
Data sharing not applicable to this article as no datasets were generated or analyzed during the current study.
References
Wells CA, Williams PD, Nichols NK, Kalise D, Poll I (2021) Reducing Transatlantic Flight Emissions by Fuel-Optimised Routing. Environ Res Lett 16(2). https://doi.org/10.1088/1748-9326/abce82
Karisch SE, Altus SS, Stojković G, Stojković M (2012) In: Barnhart, C., Smith, B. (eds.) Operations, pp. 283–383. Springer, Boston, MA. https://doi.org/10.1007/978-1-4614-1608-1_6
Alizadeh A, Uzun M, Koyuncu E, Inalhan G (2018) Optimal En-Route Trajectory Planning based on Wind Information. IFAC-PapersOnLine 51(9):180–185. https://doi.org/10.1016/j.ifacol.2018.07.030. 15th IFAC Symposium on Control in Transportation Systems CTS 2018
Rumler W, Günther T, Weißhaar U, Fricke H (2010) Flight Profile Variations due to the Spreading Practice of Cost Index Based Flight Planning. In: 4th International Conference on Research in Air Transportation, Budapest
Ng HK, Sridhar B, Grabbe S (2014) Optimizing Aircraft Trajectories with Multiple Cruise Altitudes in the Presence of Winds. Journal of Aerospace Information Systems 11(1):35–47. https://doi.org/10.2514/1.I010084
Zermelo E (1931) Über das Navigationsproblem bei ruhender oder veränderlicher Windverteilung. ZAMM 11(2):114–124. https://doi.org/10.1002/zamm.19310110205
Betts JT (2011) Practical Methods for Optimal Control and Estimation Using Nonlinear Programming (Second Edition). Siam (Society for Industrial and Applied Mathematics), Philadelphia. https://doi.org/10.1137/1.9780898718577
Dreves A, Gerdts M, Sama M, D’Ariano A (2017) Free Flight Trajectory Optimization and Generalized Nash Equilibria in Conflicting Situations. preprint
Geiger B, Horn J, DeLullo A, Niessner A, Long L (2006) Optimal Path Planning of UAVs Using Direct Collocation with Nonlinear Programming. https://doi.org/10.2514/6.2006-6199
Girardet B, Lapasset L, Delahaye D, Rabut C, Brenier Y (2013) Generating Optimal Aircraft Trajectories with Respect to Weather Conditions. In: ISIATM 2013, 2nd International Conference on Interdisciplinary Science for Innovative Air Traffic Management, Toulouse, France. https://hal-enac.archives-ouvertes.fr/hal-00867818
Girardet B, Lapasset L, Delahaye D, Rabut C (2014) Wind-optimal path planning: Application to aircraft trajectories. In: 2014 13th International Conference on Control Automation Robotics Vision (ICARCV), pp 1403–1408
Borndörfer R, Danecker F, Weiser M (2021) A Discrete-Continuous Algorithm for Free Flight Planning. Algorithms 14(1):4. https://doi.org/10.3390/a14010004
Borndörfer R, Danecker F, Weiser M (2022) A Discrete-Continuous Algorithm for Globally Optimal Free Flight Trajectory Optimization. In: D’Emidio, M., Lindner, N. (eds.) 22nd Symposium on Algorithmic Approaches for Transportation Modelling, Optimization, and Systems (ATMOS 2022). Open Access Series in Informatics (OASIcs), vol. 106, pp. 2–1213. Schloss Dagstuhl – Leibniz-Zentrum für Informatik, Dagstuhl, Germany. https://doi.org/10.4230/OASIcs.ATMOS.2022.2. https://drops.dagstuhl.de/opus/volltexte/2022/17106
Borndörfer R, Danecker F, Weiser M (2022) Error Bounds for Discrete-Continuous Shortest Path Problems with Application to Free Flight Trajectory Optimization. arXiv. https://doi.org/10.48550/ARXIV.2204.05853. https://arxiv.org/abs/2204.05853
Locatelli, M.: Simulated Annealing Algorithms for Continuous Global Optimization. In: Handbook of Global Optimization, pp. 179–229. Springer, Boston, MA (2002). https://doi.org/10.1007/978-1-4757-5362-2_6
Cassioli, A., Izzo, D., DiLorenzo, D., Locatelli, M., Schoen, F.: 5. In: Fasano, G., Pintér, J.D. (eds.) Global Optimization Approaches for Optimal Trajectory Planning, pp. 111–140. Springer, New York, NY (2013). https://doi.org/10.1007/978-1-4614-4469-5_5
Addis B, Cassioli A, Locatelli M, Schoen F (2011) A Global Optimization Method for the Design of Space Trajectories. Comput Optim Appl 48:635–652. https://doi.org/10.1007/s10589-009-9261-6
Bonyadi MR, Michalewicz Z (2017) Particle Swarm Optimization for Single Objective Continuous Space Problems: A Review. Evol Comput 25(1):1–54. https://doi.org/10.1162/EVCO_r_00180
Yen JY (1971) Finding the k Shortest Loopless Paths in a Network. Manag Sci 17(11):712–716. https://doi.org/10.1287/mnsc.17.11.712
Nocedal J, Wright SJ (2006) Numerical Optimization. Springer, New York, NY . https://doi.org/10.1007/978-0-387-40065-5
Maurer H, Zowe J (1979) First and Second-Order Necessary and Sufficient Optimality Conditions for Infinite-Dimensional Programming Problems. Math. Programming 16:98–110. https://doi.org/10.1007/BF01582096
Braess D (2013) Finite Elemente, 5th edn. Springer, Heidelberg. https://doi.org/10.1007/978-3-642-34797-9
Casas E, Tröltzsch F (2015) Second Order Optimality Conditions and Their Role in PDE Control. Jahresber Deutsch Math-Verein 117(1):3–44. https://doi.org/10.1365/s13291-014-0109-3
Funding
Open Access funding enabled and organized by Projekt DEAL. This research was funded by the DFG Research Center of Excellence MATH\(^+\) – Berlin Mathematics Research Center, Project TrU-4.
Author information
Authors and Affiliations
Contributions
Conceptualization, R.B and M.W.; methodology, F.D. and M.W.; validation, F.D.; formal analysis, F.D. and M.W.; investigation, F.D. and M.W.; resources, R.B., F.D. and M.W.; writing-original draft preparation, F.D. and M.W.; writing-review and editing, R.B.; supervision, R.B.; project administration, R.B. and M.W.; funding acquisition, R.B. and M.W. All authors have read and agreed to the published version of the manuscript.
Corresponding author
Ethics declarations
Ethics Approval
Not applicable
Consent to Participate
Not applicable
Consent for Publication
We confirm that all authors agree with the submission of this manuscript to Public Transport Optimization: From Theory to Practice.
Competing Interests
The authors declare no competing interests.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix
Appendix
1.1 A.1 Global bounds
The derivative \(f=t_\tau\) of parametrized time as defined in (3) consists of two terms, the tailwind term
and the length term
At each time \(\tau\), we obtain
The directional derivatives of g in direction \({\delta \xi }\) and \({\Delta \xi }\in {\delta X}\) read
For the tailwind term, we consider
which is bounded by
The second directional derivatives is
and in particular
which yields
and
respectively.
Finally, the third directional derivative is
which is bounded by
Before we turn to the length term \(f_2\), we first consider the term
with
We also note that
Then
which is bounded by
The second derivative is
and in particular
which yields
and
respectively. The third derivative is
which is bounded by
For the length term \(f_2 = g^{-1} \sqrt{F}\), we thus obtain
which is bounded by
The second derivative is
and in particular
which yields
and
The third derivative is
which is bounded by
Lemma 16
Let \(\Vert w(p)\Vert \le {\overline{c}}_0 \le {\overline{v}}/\sqrt{5}\), \(\Vert w_x(p)\Vert \le {\overline{c}}_1\), \(\Vert w_{xx}(p)\Vert \le {\overline{c}}_2\), and \(\Vert w_{xxx}(p)\Vert \le {\overline{c}}_3\) for every \(p\in \Omega\). Moreover let \({\underline{v}}^2:= {\overline{v}}^2 - {\overline{c}}_0^2\) and \({\overline{\overline{v}}}^2:= {\overline{v}}^2 + {\overline{c}}_0^2\). Then, for any \(\xi \in X\), the third directional derivative of f as given in (3) is bounded by
with
Proof
We obtain f by adding \(f_1\) and \(f_2\). The third derivative of f can thus be bounded using (65), (82), and the triangle inequality.
With \(\frac{{\overline{c}}_0}{{\overline{v}}} \le \frac{1}{\sqrt{5}}\), we note that
and obtain
Rounding up the values yields the bound
Lemma 17
Let \(\Vert w(p)\Vert \le {\overline{c}}_0 \le {\overline{v}}/\sqrt{5}\), \(\Vert w_x(p)\Vert \le {\overline{c}}_1\), and \(\Vert w_{xx}(p)\Vert \le {\overline{c}}_2\) for every \(p\in \Omega\). Moreover let \({\underline{v}}^2:= {\overline{v}}^2 - {\overline{c}}_0^2\) and \({\overline{\overline{v}}}^2:= {\overline{v}}^2 + {\overline{c}}_0^2\). Then, for any \(\xi \in X\), the second directional derivative of f as given in (3) is bounded by
with
Proof
We obtain f by adding \(f_1\) and \(f_2\). The second derivative of f can thus be bounded using (62), (79), and the triangle inequality.
With \(\frac{{\overline{c}}_0}{{\overline{v}}} \le \frac{1}{\sqrt{5}}\), we note that
and obtain
Rounding up the values yields the bound
1.2 A.2 Bounds in a neighborhood of a minimizer
Below we derive bounds that hold in a \(L^\infty\)-neighborhood of a global minimizer. Let \(x^{\star \star } = (z^{\star \star }, \lambda ^{\star \star })\) be a global minimizer of (6) and the corresponding Lagrange multipliers. Moreover, let \(x_1,x_2 \in \mathcal {B}(x^{\star \star },R)\) and define \(\Delta x:= x_2-x_1\). Then it holds that \(\Vert \Delta x\Vert _{Y^{\infty }} \le 2R\) and consequently
Let \(\Vert w\Vert _{L^{\infty }(\Omega )} \le {\overline{c}}_0\), \(\Vert w_x\Vert _{L^{\infty }(\Omega )} \le {\overline{c}}_1\), \(\Vert w_{xx}\Vert _{L^{\infty }(\Omega )} \le {\overline{c}}_2\), and \(\Vert w_{xxx}\Vert _{L^{\infty }(\Omega )} \le {\overline{c}}_3\), then the following bounds hold,
Moreover, we show that
Furthermore, with F as given in (66), and (28) we get
and
This yields
For \(f_1\) as defined in (48), we obtain
Using the bounds from above we finally obtain
with
For \(f_2\) as defined in (50) we obtain
Using the bounds from above, this yields
with
Lemma 18
Let \(x^{\star \star } = (z^{\star \star }, \lambda ^{\star \star })\) be a global minimizer of (6) and the corresponding Lagrange multipliers. Moreover, let \(x_1,x_2 \in \mathcal {B}(x^{\star \star }, R)\) and define \(\Delta x:= x_2-x_1\). Then there is a \(\hat{B} < \infty\) such that
Proof
with \(\hat{B} = \max \{ \hat{\beta }_1, \hat{\beta }_2 \}\).
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Borndörfer, R., Danecker, F. & Weiser, M. Newton’s Method for Global Free Flight Trajectory Optimization. Oper. Res. Forum 4, 63 (2023). https://doi.org/10.1007/s43069-023-00238-z
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s43069-023-00238-z
Keywords
- Shortest path
- Flight planning
- Free flight
- Optimal control
- Discrete optimization
- Global optimization
- Newton’s method