arXiv:1910.06452v5 [cs.GT] 7 Sep 2021
Transcript of arXiv:1910.06452v5 [cs.GT] 7 Sep 2021
When Nash meets Stackelberg
Margarida Carvalho1,2, Gabriele Dragotto2, Felipe Feijoo4, Andrea Lodi2.3, and SriramSankaranarayanan5
1 Universite de Montreal [email protected] CERC Data Science, Polytechnique Montrealgabriele.dragotto, [email protected]
3 Jacobs Technion-Cornell Institute, Cornell Tech, New York City4 Pontifica Universidad Catolica de Valparaıso [email protected]
5 Production and Quantitative Methods, Indian Institute of Management, Ahmedabad [email protected]
Abstract. Motivated by international energy trade between countries with profit-maximizing domesticproducers, we analyze Nash games played among Stackelberg games leaders (NASP). In particular, wefocus on NASPs where each leader program is a linear bilevel with quadratic convex followers, and weassume the standard optimistic version of such bilevels. We prove it is both Σp
2 -hard to decide if thegame has a pure-strategy (PNE) or a mixed-strategy Nash equilibrium (MNE). We provide a finitealgorithm that computes exact MNEs for NASPs when there is at least one or returns a certificate ifno MNE exists. To enhance computational speed, we introduce an inner approximation hierarchy thatincreasingly grows the description of each Stackelberg leader feasible region. Furthermore, we extend thealgorithmic framework to retrieve a PNE if one exists specifically. Finally, we provide computationaltests on a range of NASPs instances inspired by international energy trades.
Keywords: Game Theory, Algorithmic Game Theory, Stackelberg Game, Nash Game, EquilibriumProblems with Equilibrium Constraints, Mixed Integer Programming
1 Games, definitions, and overview
Game theoretical frameworks model complex interactions among agents and are widely employed for real-world applications. Their effectiveness relies on two key ingredients. First, their modeling capabilities forthe specific field of application and the ease of interpretability of such models. Secondly, the power andefficiency of the underlying algorithmic arsenal available to solve these models. In this paper, we study a classof non-cooperative, simultaneous games between the leaders (i.e., the first-level players) of bilevel programswith an optimistic followers’ response. In other words, Stackelberg games’ leaders are playing a Nash gameamong themselves with complete information. We call such problems Nash Games among Stackelberg Leaders( NASPs), schematically represented in Figure 1. NASPs are part of the well-known family of equilibriumproblems with equilibrium constraints ( EPECs), which has a wide variety of applications in energy markets. Aconcise representation of an elementary (or trivial, as more formally defined in Definition 7) NASP between aLatin Stackelberg game and a Greek Stackelberg game is given byLatin Leader
minx,y
: cTx+ dT y +
(G
(ξχ
))T (xy
)(1a)
subject to Ax+By ≤ b (1b)
y ∈ arg miny
fT y : Qy ≤ g − Px
(1c)
Greek Leader
minξ,χ
: αT ξ + βTχ+
(Γ
(xy
))T (ξχ
)(1d)
arX
iv:1
910.
0645
2v5
[cs
.GT
] 7
Sep
202
1
Leader1
F 11 F 1
f1
Leader`
F 1` F `
f`
Fig. 1: A schematic representation of a NASP . The vertical arrows are Stackelberg interactions (i.e, sequentialdecisions), while the horizontal ones are Nash interactions (i.e, simultaneous decisions).
subject to Φξ + Ψχ ≤ ρ (1e)
χ ∈ arg minχ
φTχ : Ωφ ≤ γ −Πξ
. (1f)
The interaction among the Stackelberg leaders happens within their objective function, as in (1a) and (1d).On the other hand, within each Stackelberg game, each leader anticipates the reaction of their distinctfollowers, each of which solves a lower-level parametric linear program. NASPs can be extended (i) tohandle more than two leaders, (ii) to handle more than one follower per leader, (iii) to have the follow-ers of each leader interact in a Nash game, and (iv) to enforce each follower to solve a convex quadraticprogram as opposed to linear programs in (1c) and (1f). A dominant solution concept is the one of Nashequilibrium — namely when each player cannot profitably and unilaterally deviate from the prescribed strategy.
Applications. NASPs, in their full generality, could solve a wide range of problems. We outline three differentpotential applications related to energy, vaccines, and insurances. In this work, we are primarily motivated tomodel international energy markets with climate change-aware regulatory authorities with profit-maximizingdomestic energy producers, and we provide a game-theoretic framework to analyze this problem. In thisgame, energy producers – namely the Stackelberg followers – compete in the domestic market and are usuallysubject to restrictions in the form of tax and caps from the regulatory authorities. The regulatory authorities– namely the Stackelberg leaders – negotiate environmental-conscious agreements for energy trade, thusengaging in a Nash game. The theoretical abstraction NASP models this problem and provides a generalframework to analyze games, in and outside the field of energy, when there are multiple Stackelberg leaders,each with their set of followers, playing a Nash game with each other.Similarly – yet in a different context – NASPs can model a complex drug trade system. For instance, at thetime of writing, the COVID-19 vaccine production and trade situation pose severe threats to the world’simmunization programs and may lessen inter-country cooperation with the so-called vaccine nationalism(Weintraub et al. 2020). In several cases, countries threatened and successfully blocked vaccine exports whilealso imposing strict regulations on indigenous producers (Fleming et al. 2021, Boffey 2021a,b). In this scenario,the homogeneous good would be the vaccine, and in analogy to emission factors, we would see efficacyproperties. Countries act as Stackelberg leaders, regulating vaccines’ trade and incentivizing indigenousproducers (followers). Further, the leaders’ objectives could model a wide variety of tactical requirements (e.g.,prioritize the production of some doses reserved for vulnerable classes, incentivize the exports of prioritizeddoses to neighboring countries, prioritize more effective vaccines).
2
Finally, as a third example, we draw attention to the insurance business. Users of a given good may be subjectto disruptions (e.g., cyberattacks for energy generators) and may need to contract insurance services (Gordonand Loeb 2002, Cashell et al. 2004). Insurers – namely Stackelberg leaders – provide such services at a cost totheir clients – or followers. The NASP framework extends this hierarchical model to a multi-insurer setting,introducing a mechanism of re-insurance. In plain English, the leaders mutually protect their insurances’portfolios, shielding them against large-scale disruptions (e.g., natural disasters). With these motivatingexamples, we now detail our primary contributions.
Primary Contributions
In the first place, we characterize the computational complexity of deciding if a given instance of NASP hasa pure-strategy Nash equilibrium (PNE ). Even with restrictive assumptions – such as single follower andbounded strategy sets for all players – we show it is Σp
2 -hard to decide if the instance has a PNE . Namely,even with oracle access to solve NP-hard problems instantaneously, there would be exponentially many callsto such an oracle to decide the existence of PNE for a given instance of NASP . In other words, withoutsubstantial consequences in complexity theory, this translates to a bound of Ω(22n
) elementary operationsrequired to solve the problem, where n is the size of the representation of the corresponding decision problem.This is quite surprising since, in most literature cases, one can either prove that all games in a consideredcategory have a PNE or prove sufficiently fast that a given instance has no PNE . Second, we consider thecomputational complexity of deciding the existence of mixed-strategy Nash equilibrium (MNE ) for NASPs.We demonstrate that with exactly one follower for each leader and boundedness in every player’s problem, anMNE always exists (Corollary 1). However, even if one of the leaders has an unbounded feasible set, it isagain Σp
2 -hard to decide the existence of an MNE .Third, given these lower bounds to computationally find PNE or MNE for NASPs, we provide a finite-time
algorithm to do so. It retrieves an MNE for an instance of NASP when it exists and provides a (doubleexponentially-large) proof of infeasibility when an MNE does not exist. To the best of our knowledge, this isthe first algorithm to identify MNE or PNE for a game of this type. Fourth, we provide an enhancementto the algorithm to exclusively seek PNEs, or provide proof of infeasibility. This is the case of interest ifmixed-strategies are not implementable in practice. Fifth, we provide another enhancement to the algorithmsto find MNEs and PNE s, with an iterative inner-approximation procedure that proves to be considerablyfaster in practice. We also remark that the negative results (Σp
2 -hard complexity) are for the easier versionof the problem (the latter defined trivial NASP), and our positive algorithmic results are, on the contrary,for the harder version of the problem with multiple followers. Besides, we also present several observations,for instance, Remarks 1 and 3, which enlight on equilibria for Nash games where players solve non-convexoptimization problems.
We believe that the above contributions, both from the complexity and algorithmic (computational) sidesestablish a solid benchmark against which future progress can be measured.
Literature review.
Nash games gained popularity with the Nobel awarded papers of Nash (1951, 1950). Nash proven the existenceof the so-called Nash equilibrium for games with a finite number of players and a finite number of strategies.By definition, these equilibrium strategies ask that no player has an incentive to unilaterally deviate fromthe prescribed strategy. Generally, we distinguish between the pure strategy Nash equilibrium (PNE ) andthe mixed strategy one (MNE ). The latter generalizes the pure one since each strategy in the support ofthe equilibrium has an associated probability of being played. The Nash equilibrium concept extends togames where players have an uncountable set of strategies. From an application perspective, Nash Gamesare extensively adopted for modeling interactions within economic markets. For instance, gas market bilevelformulations usually involve players solving convex optimization problems parametrized in other playersvariables (Egging et al. 2010, Feijoo et al. 2016, Sankaranarayanan et al. 2018, Feijoo et al. 2018, Holz et al.2008, Egging et al. 2008, Stein and Sudermann-Merx 2018). On the other side, the cross-border kidney
3
exchange model (Carvalho et al. 2017), competitive lot-sizing models (Li and Meissner 2011, Carvalho et al.2018b), and the fixed charge transportation model (Sagratella et al. 2019) feature players solving non-convexproblems.
In contrast with Nash games, sequential ones partition the players into different groups, and each groupdecides in a round – or level. If the rounds are two, then the game is known as Stackelberg game (Stackelberg1934, Candler and Norto 1977). Here, the agents playing in the first round are the leaders, while the onesplaying afterward are called followers. When each Stackelberg player solves an optimization problem, then wehave a bilevel program. In general, bilevel formulations can model interactions where leaders have specificadvantages over the followers, such as government taxing companies. Indeed, bilevel formulations allure anourished community of researchers. Bard et al. (1998, 2000) model tax credits strategies in the context ofbiofuel production, and Brotcorne et al. (2008), Labbe and Violin (2013) create bilevel pricing problems.Hobbs et al. (2000), Gabriel and Leuthold (2010), Feijoo and Das (2014) model pricing and environmentalpolicies for energy markets, where power generators are leaders, and network operators are followers.
When multiple leaders – each with possibly multiple followers – seek for an equilibrium between eachother, we fall into the category of EPEC s. Thereby leaders often have a common set of followers, andthe equilibrium of interest is PNE . Sherali (1984) introduced EPEC s where both leaders and followersproduce a homogeneous commodity, and followers adopt a reaction curve. Gabriel et al. (2012) providesa Gauss-Seidel iteration technique to find PNEs for a restricted class of EPEC s, where followers fromdistinct leaders can interact. Ralph and Smeers (2006), and Hu and Ralph (2007) extend the analysis on theexistence of a PNE to specialized classes of EPEC s arising in electricity markets. Leyffer and Munson (2010)introduces a weaker solution concept based on a nonlinear programming reformulation. DeMiguel and Xu(2009) craft the concept of stochastic multi-leader Stackelberg-Nash-Cournot equilibrium for a particularform of investment-production interaction between the players. More recently, Kulkarni and Shanbhag (2014,2015) considered EPEC s with shared constraints, presenting solution concepts and algorithms starting fromthe potentiality of players’ objectives.
Complexity of Equilibria. As previously mentioned, Nash (1950, 1951) proven that a Nash equilibrium forfinite games always exists, and thus the associated decision problem is trivial. However, since the proof isnon-constructive, it already unveils the difficulty of computing an equilibrium. Indeed, even for two-playersfinite games in strategic form, the problem of determining an equilibrium is PPAD-complete (Chen and Deng2006). Furthermore, even for games where equilibria are guaranteed to exist, many variations of associateddecision problems are known to be NP-complete (Gilboa and Zemel 1989). A few illustrative examples arethe existence of two equilibria or the existence of an equilibrium where a player’s payoff exceeds a giventhreshold. Besides, Carvalho et al. (2018a, 2020) proved the existence of PNE and MNE for games whereplayers solve parametrized non-convex problems to be Σp
2 -hard. Under this setting, if players’ strategies arebounded, then an MNE always exists. For congestion games, another widely studied class of Nash games,PNE s always exist due to their potential nature (Rosenthal 1973). Del Pia et al. (2017) focus on congestionsgames where totally unimodular matrices describe the players’ strategies. Within this context, the authorsprove that if players have the same feasible set of strategies, a PNE can be computed in polynomial time. Inany other case, the problem is PLS-complete. For what concerns Stackelberg games’ complexity, the seminalresult of Jeroslow (1985) enlightens the matter. It proves that sequential games’ computational complexityrises one layer up in the polynomial hierarchy for every additional round, even for linear problems. Thereupon,the classification of the computational complexity for NASPs becomes almost natural.
Paper Organization. We organize the manuscript as follows. Section 2 provides definitions and restates someknown results. Section 3 provides the complexity results regarding NASPs. Section 4 presents an algorithmto find MNE for NASP , proves its finiteness and correctness. Section 5 builds on top of the developedalgorithm by extending it with an inner approximation hierarchy and introduces a heuristic for computingPNE . Section 6 presents computational tests, and, finally, Section 7 draws conclusions.
4
2 Preliminaries
In this section, we provide definitions, notations and recall some known results in the context of polyhedraltheory, Nash games, and Stackelberg games.
2.1 Definitions
Nash Games. When players decide simultaneously, and with complete information, we have a Nash Game.As a standard notation in game theory, let the operator (·)−i denote (·) except i.
Definition 1 (Nash games). A Nash game P among n players is a finite tuple of optimization problemsP =
(P 1, . . . , Pn
), where each P i is the problem of the ithplayer. Simultaneously, each player i solves an
optimization problem of the form minxi∈Rni f i(xi;x−i) : xi ∈ Fi, where f i and Fi are their objectivefunction and the feasible set, respectively.
Moreover, we can further characterize a Nash game as (i) simple if, for every player i and for somepositive semidefinite matrix Qi, and ci, Ci of appropriate dimensions, the objective function is in the form
of f i(xi;x−i) = 12x
iTQixi +(ci + Cix−i
)xi, (ii) linear, if Qi = 0 for all i, namely each leader has a linear
objective function, (iii) facile, if the game is simple, and Fi is a polyhedron for all i ∈ [n].
Definition 2 (Simple parameterization). An optimization problem in y has a simple parameterizationwith respect to x ∈ Rn` if the problem is in the form of miny∈Rnf f(y) + (Cx)T y : y ∈ F , Ax + By ≤ b,where f : Rnf → R, and C, A, B, b are matrices and vectors of appropriate dimensions, and F ⊆ Rnf .
A Nash game P = (P 1, . . . , Pn) has a simple parameterization with respect to x ∈ Rn` if each optimizationproblem P 1(x), . . . , Pn(x) has a simple parameterization with respect to x.
Definition 3 (Mixed and Pure-strategy Nash equilibria). Let ν = (ν1, . . . , νn) where νi is a Borelprobability distribution on Fi with finite support. Then, ν is a MNE if – for all i ∈ [n] and xi ∈ Fi –E(f(νi, ν−i)) ≤ E(f(xi, ν−i)) holds. If all the distributions have a singleton support, then the set of strategiesis referred to as PNE.
PNE is a strong notion of equilibrium, and even relatively trivial games — for example, rock-paper-scissors —do not possess one. In contrast, an MNE always exists for finite games (Nash 1950, 1951).
Stackelberg Games. A Stackelberg game is a multi-level game with 2 rounds of decisions. First, the leaderdecides, optimizing their objective subject to some constraints. Subsequently, the followers decide, with theirobjective and constraints now depending upon the leader’s decision (Candler and Townsley 1982).
Definition 4 (Stackelberg game). Let P (x) be a Nash game with a simple parametrization with respectto x, SOL(P (x)) denotes its solution set, and f : Rn`+nf → R. Then, a Stackelberg game is an optimizationproblem of the form min
x∈Rn` ;y∈Rnff(x, y) : (x, y) ∈ F , y ∈ SOL(P (x)).
In a Stackelberg game, the set SOL(P (x)) is parametrized given the leader’s strategy x. Namely, given anupper-level strategy x, the followers should play optimally. Theoretically, each Stackelberg game’s solution isthen a subgame perfect Nash-Equilibrium (SPNE). For the purposes of this manuscript, we only considerSPNEs. The previous definition implies the Stackelberg game to be optimistic. Namely, if the game hasmultiple optimal solutions SOL(P (x)), then y takes the value among SOL(P (x)) benefitting the leader themost. If P (x) is an optimization problem (i.e., one follower), then the optimistic assumption is natural aswe could state that the leader incentivizes the follower by an arbitrarily small amount to choose the mostfavorable solution. However, this could be a strict restriction if there are multiple followers and the followers’equilibrium is not unique. As commented before, the optimistic assumption on the bilevel solution selection isnot restrictive here, because despite having multiple followers, their equilibrium is unique. Such uniquenessfollows from the fact that the followers play a Nash-Cournot game with strictly convex cost functions.
5
Definition 5 (Simple Stackelberg game). A Stackelberg game P is simple if P (x) is a facile Nash gamewith a simple parameterization with respect to the upper-level variables x, F is a polyhedron, and f(x, y) is alinear function.
Definition 6 (NASP). A NASP is a linear Nash game N = (P 1, . . . , P k) where for each i, P i(xi) is asimple Stackelberg game.
Combining Definitions 1, 2 and 4 to 6, a NASP refers to the following game. There is a set L of players calledthe leaders and another set F of players called the followers. Each follower f ∈ F has a unique leader `(f) ∈ L,such that the objective function and the feasible set of the follower f depends only upon the decision variablesof `(f) and other followers f ′ ∈ F such that `(f ′) = `(f). In other words, only the followers of the sameleader interact with each other, and they only interact with their leader. We assume all parameters of theplayers are common information. First, with complete anticipation of their followers’ behavior, simultaneously,every leader ` ∈ L chooses their decision to maximize their utility. Then, every follower f observes theirrespective leader’s (`(f)’s) decision and, simultaneously, every follower chooses their decision, maximizingtheir utility. We assume an optimistic behavior from the followers in the sense that – if there are multipleoptimal strategies for the followers over which they are indifferent – they will choose the strategy whichbenefits their leader the most.
Definition 7 (Trivial NASP). A trivial NASP is a NASP where k = 2, and P 1 and P 2 are simple bilevelgames whose lower levels are linear programs with a simple parameterization with respect to the upper-levelvariables.
The additional assumptions holding on a trivial NASP (as of Definition 7) compared to a general NASP (asof Definition 6) are seemingly strong. We require that each leader has precisely one follower – as opposed tofinitely many followers – and that each follower solves a linear program – as opposed to a quadratic program –with a simple parameterization with respect to the upper-level variables. The game between the Latin andGreek leaders presented in (1) is an example of trivial NASP . Finally, in NASPs with lower-level facile NashGames, the feasible region for the followers are convex, and the leaders’ objective functions are convex in xi.As a consequence, the existence of a PNE is guaranteed whenever the feasible regions are compact (Debreu1952, Glicksberg 1952, Fan 1952). Therefore, one can restrict the search of equilibria to PNEs among thefollowers despite looking for both MNE s and PNE s among the leaders.
Typically, within the optimization literature, Nash games are reformulated as linear complementarityproblems (LCPs). This reformulation leverages the complementarity conditions (KKT ) of all the players inthe game. LCPs have a rich theoretical basis (Facchinei and Pang 2015b,a, Cottle et al. 2009), and can beformulated as mixed-integer programs (MIPs). Following the usual notation, let operator x ⊥ y be equivalentto xT y = 0.
Definition 8 (Linear complementarity problem). Given M ∈ Rn×n, q ∈ Rn, the linear complemen-tarity problem ( LCP) asks to find a x ∈ Rn so that 0 ≤ x ⊥Mx+ q ≥ 0, or to show that no such x exists.We denote as feasible set induced by the LCP, the set of all x satisfying the condition of the LCP.
Simplifying assumptions and limitations. Here, we sum up the assumptions made for the entirity of thispaper. First, a NASP is a game among leaders of Stackelberg games, each of who have a personal set offollowers. In other words, actions of the followers of a certain leader do not affect another leader or followersof another leader. Second, we assume optimistic behavior by the leader. That is, given a leader’s decision,should there be multiple equilibria for the followers, the leader will choose the most favorable equilibriumfrom its standpoint. This also has consequences for the upcoming results about absence of equilibria. Whenwe state that a NASP does not have an equilibrium (PNE or MNE), our claim is as follows: should theleaders always pick the most favorable lower-level equilibrium, no equilibrium exists among the leaders. Thisindeed means, if the leader does not pick the most favorable equilibrium among the followers, an equilibriumamong the leaders might exist. In this sense, we consider non-optimistic equilibrium selection by the leadersto be beyond the scope of this work.
6
2.2 Known results.
Cottle et al. (2009) proven that a facile Nash game can be solved as an LCP .
Theorem 1 (Cottle et al. (2009)). Let P be a facile Nash game. Then, there exist M, q such that everysolution to the LCP defined by M, q is a PNE for P and every PNE of P solves the LCP.
Basu et al. (2019), with Theorem 2, provide an extended formulation for the feasible region of a simpleStackelberg game. This result is a critical ingredient of our contribution since it enables us to provide apolyhedral characterization NASPs.
Theorem 2 (Basu et al. (2019)). Let S be the feasible set of a simple Stackelberg game. Then, S isa finite union of polyhedra. Conversely, let S be a finite union of polyhedra. Then, there exists a simpleStackelberg game with P (x) containing exactly 1 player such that the feasible region of the simple Stackelberggame provides an extended formulation of S.
Finally, the celebrated Theorem 3 from Balas (1985) allows us to retrieve the closure of the convex hull forthe union of a finite set of polyhedra.
Theorem 3 (Balas (1985)). Given k polyhedra Si = x ∈ Rn : Aix ≤ bi for i = 1, . . . k, then
cl conv(⋃ki=1 Si) is given by the set x ∈ Rn : ∃(x1, . . . , xk, δ) ∈ (Rn)k × Rk : x ∈ Aixi ≤ δib
i,∑kw=1 x
w =
x,∑kw=1 δw = 1, δi ≥ 0,∀i ∈ [k]
3 Hardness of finding a Nash equilibrium
In what follows, we characterize the computational complexity of NASPs. We formalize the intuitionstemming from Jeroslow (1985) with a reduction from the SUBSET SUM INTERVAL problem. The main resultsare summarized below.
Theorem 4. It is Σp2 -hard to decide if a trivial NASP has a PNE.
Corollary 1. If each player’s feasible set in a trivial NASP is a bounded set, an MNE exists.
Theorem 5. It is Σp2 -hard to decide if a trivial NASP has an MNE.
In what follows, we will provide the proof of Theorems 4 and 5. First, we formally introduce the SUBSET
SUM INTERVAL.
Definition 9 (SUBSET SUM INTERVAL). Given q1, . . . , qk, p, t, k ∈ Z+, with none of them equal to zero, andlog2(t− p) ≤ k, does there exist a s ∈ Z : p ≤ s < t, so that for all I ⊆ 1, 2, . . . , k then
∑i∈I qi 6= s.
In other words, we seek – within an interval of integers – for a number s that cannot be expressed as a sumof a subset of q1, . . . , qk or alternatively show that no such s exists. Here, t− p can be chosen as a powerof 2. For instance, we may ask if there exist an r in ∈ Z+ such that 2r = t− p. Eggermont and Woeginger(2013) proven that, given r in Z+ such that t− p = 2r, the problem is Σp
2 -hard.
Theorem 6 (Eggermont and Woeginger (2013)). Given that there exists r ∈ Z+ such that t− p = 2r,SUBSET SUM INTERVAL is Σp
2 hard.
Proof. Proof of Theorem 4. To show the hardness of NASP, we will rewrite SUBSET SUM INTERVAL as atrivial NASP of comparable size. Then, appealing to Theorem 6, we establish the hardness of a trivial NASP .Finally, we claim that NASP is only a generalization of trivial NASP , which could not be any easier.
Consider a trivial NASP as of in Definition 7. For the sake of clarity, we call the two Stackelberg gamesassociated with the trivial NASP the Latin, and Greek game, respectively. The decision variables of the Latingame’s leader are x, and their follower controls y variables. Similarly, the decision variables of the Greekgame are ξ, and χ for their follower. As for the SUBSET SUM INTERVAL, we stick to the notation introducedin Definition 9.
Let b1, . . . , br ∈ 0, 1 as the unique r-bit binary representation of s− p: for instance, biri=1 satisfies
s− p =∑ri=1 bi2
i−1. Then, let P = k + 2r, Q =∑ki=1 qi, and T = t− 1 + rQ, where both can be computed
in polynomial time with respect to the data in SUBSET SUM INTERVAL.
7
The Latin Game.
maxx0,x1,...,x2P
∈Ry0,y1,...,y2P
∈R
: (T − 1)ξ0x0 +
k∑i=1
qiξixP+i +Q
P∑i=k+1
ξixP+i (2a)
subject to xi = 0 i = 1, . . . , k (2b)
yi ≥ 0 i = 1, . . . , 2P (2c)
xi ≥ 0 i = 1, . . . , 2P (2d)
P∑i=k+1
xi ≤ r (2e)
xi + xP+i ≤ 1 i = 1, . . . , P (2f)
x0 + xP+i ≤ 1 i = 1, . . . , P (2g)
(y0, . . . , y2P ) ∈ arg miny
2P∑i=0
yi :yi ≥ −xiyi ≥ x1 − 1
∀ i = 0, . . . , 2P
(2h)
The Greek Game.
maxξ0,ξ1,...,ξP∈R
χ0,...,ξP∈R
: (T − 1)ξ0 +
k∑i=1
qiξi(1− xP+i) +Q
P∑i=k+1
ξi(1− xi − xP+i)
+
k+r∑i=k+1
2i−k−1ξi(1− xi − xP+i)
−P∑
i=k+1
T (xiξi + (1− xi)(1− ξi − ξ0)) (2i)
subject to ξi ≥ 0 ∀ i = 0, . . . , P (2j)
ξi ≤ 1 ∀ i = 0, . . . , P (2k)
χi ≥ 0 ∀ i = 0, . . . , P (2l)
P∑i=k+1
ξi + rξ0 ≥ r (2m)
T ≥ Tξ0 +
k∑i=1
qiξi +Q
P∑i=k+1
ξi +
k+r∑i=k+1
2i−k−1ξi (2n)
(χ0, . . . , χP ) ∈ arg minχ
P∑i=0
χi :χi ≥ −ξiχi ≥ ξi − 1
∀ i = 0, . . . , 2P
(2o)
We claim the game in (2) has a PNE , if and only if the SUBSET SUM INTERVAL instance has a decision YES.
Claim 1 The game defined in (2) is a trivial NASP.
Claim 2 The region in the space of x defined by (2c) and (2h) is the Cartesian product of (xi : xi ≤ 0 ∪ xi : xi ≥ 0),for i = 0, . . . , 2P . Similarly, the region in the space of ξ defined by (2l) and (2o) is the Cartesian product of(ξi : ξi ≤ 0 ∪ ξi : ξi ≥ 0), for i = 0, . . . , P .
8
We refer the reader to the electronic companion for the proofs of Claims 1 and 2.
Claim 3 If((x, y), (ξ, χ)
)is a PNE for (2), then ξ0 6= 0.
Proof (Proof of Claim 3). First, observe that ξ 6= 0, since setting ξ0 = 1 is a feasible profitable deviation forthe Greek leader, regardless of the Latin leader’s decision. Suppose ξ0 = 0 and for some ∅ 6= L ⊆ 1, . . . , P,ξ` 6= 0. Note that the Latin leader has no incentive to keep x0 = 1, which forces an objective value of 0.Instead, it can choose x0 = 0, and xP+` = 1 for all ` ∈ L and any feasible value for xP+` for ` ∈ 1, . . . , P\L.One can check that this is feasible and optimal for the the Latin leader, given ξ0 = 0. This also means thatthe Greek leader’s objective is 0, as each of the summands in their objective vanishes, and ξ0 = 0 makes thefirst term vanish. Hence, this cannot be a Nash equilibrium since the Greek leader has a profitable deviationby setting ξ0 = 1 and ξi = 0 for all i 6= 0, which is feasible and yields an objective value of T − 1 > 0.
Claim 4 If SUBSET SUM INTERVAL has decision YES, then (2) has a PNE.
Proof (Proof of Claim 4). Suppose there exists s ∈ Z+ such that p ≤ s ≤ t− 1, and for all I ⊆ 1, . . . , k,∑i∈I qi 6= s. Also, recall the unique r-bit binary representation of s− p, namely b1, . . . , br ∈ 0, 1. Consider
the following strategy:
x0 = 1 (3a)
xi = 0 ∀ i = 1, . . . , k (3b)
xi = bi−k ∀ i = k + 1, . . . , k + r (3c)
xi = 1− bi−k−r ∀ i = k + r + 1, . . . , P = k + 2r (3d)
xi = 0 ∀ i = P + 1, . . . , 2P (3e)
yi = 0 ∀ i = 0, . . . , 2P (3f)
ξ0 = 1 (3g)
ξi = 0 ∀ i = 1, . . . , P (3h)
χi = 0 ∀ i = 1, . . . , P (3i)
It is easy to check that the strategy in (3) is feasible. Given ξ, observe that the strategy is optimal for theLatin leader as follows. Due to the choice ξi = 0 for i 6= 0, all but the first term of the Latin leader vanish.The largest value the first term can take corresponds to x0 = 1. The remaining terms do not affect the Latinleader’s objective, as long as they are feasible.
For what concerns the Greek leader, the current objective is T − 1. We show there exist no deviationwhich can improve their objective. With ξ0 = 1, clearly no other deviation is feasible. Consider the deviationξ0 = 0: with such strategy the first term in the objective vanishes. Let M = i ∈ k+ 1, . . . , k+ 2r : xi = 1.Observe that |M | = r, and let L = k+ 1, . . . , k+ 2r \M . Notice that we require ξ` = 1 for ` ∈ L, otherwisethe fifth term in the objective would be a large negative quantity. Hence, the objective would not exceedthe value of T − 1. With such a choice of ξ` for ` ∈ L, the fifth term in the objective evaluates to 0, and thefourth term evaluates to
∑`∈L 2`−k−i =
∑k+ri=k+1(1− bi−k)2i−k−1 = 2r − 1 + p− s = t− 1− s. Therefore, the
objective value is t−1 + rQ−s. However, since it is a YES instance of SUBSET SUM INTERVAL, the deficit s inthe objective value can never be made up by any choice of ξi for i = 1, . . . , k and by making the second termequal to s. If such ξi are chosen to exceed s, then (2n) is violated if it is strictly less than s, and the objectivecannot exceed T − 1. Hence, this is no longer a valid deviation. Thus (3) is indeed a Nash equilibrium. ut
Claim 5 If SUBSET SUM INTERVAL has decision NO, then (2) has no PNE.
Proof (Proof of Claim 5). We prove the result by contradiction. In orter to establish the latter, assume that theSUBSET SUM INTERVAL instance has an answer NO, and there exists a PNE
((x, y), (ξ, χ)
)for (2), with ξ0 = 1.
From Claims 2 and 3, any PNE necessarily has ξ0 = 1. From (2n), ξ0 = 1 enforces that ξi = 0 for i = 1, . . . , T ,and hence the Greek leader has an objective value of T − 1. Therefore, with ξ =
(1 0 . . . 0
), observe that
the Latin leader’s objective is (T − 1)x0. Thus, we necessarily have x0 = 1. From (2g), we deduce xP+i = 0
9
for i = 1, . . . , P , while from (2e) we obtain xi ≤ rr+1 for i = 1, . . . , k. The only value of xi that satisfies this
condition along with (2h) is xi = 0 for i = 1, . . . , k That only leaves xi for i = k + 1, . . . , k + 2r = P .We can now show that – for any value of xi – the Greek leader has a profitable deviation. Namely, it
can get an objective strictly greater than T − 1. Let M = i ∈ k + 1, . . . , k + 2r : xi = 0. From (2e), wehave |M | ≥ r. We choose some L ⊆ M such that |L| = r, and for i ∈ L, we set ξi = 1. Since |L| = r, andL ⊆M , the third term in the Greek leader’s objective evaluates to rQ. The fourth term is in between 0 and2r − 1, and the fifth term vanishes. Keeping in mind that ξ0 = 0, the objective now evaluates to a number
between∑ki=1 qiξi + rQ and
∑ki=1 qiξi + rQ+ 2r − 1. In other words, the objective is T − s+
∑ki=1 qiξi and
p ≤ s ≤ t− 1. Since this is a NO instance of SUBSET SUM INTERVAL, ∃I ⊆ 1, . . . , k such that∑i∈I qiξi = s.
Set ξi = 1 if i ∈ I, and ξi = 0 if i ∈ 1, . . . , k \ I. This is feasible, and makes the objective value equal to T ,which is a profitable deviation from T − 1. Therefore
((x, y), (ξ, χ)
)is not a Nash equilibrium. ut
From Theorem 4, we have a direct implication of Corollary 2.
Corollary 2. Consider a linear Nash Game N = (P 1, . . . , Pn) where each P i is an MIP . It is Σp2 -hard to
decide if N has a PNE.
Proof (Proof of Corollary 2). Bounded and continuous bilevel programs can be reformulated as boundedinteger programs of polynomial-size (Basu et al. 2019). The Greek and the Latin leaders’ problems defined in(2) are bounded bilevel programs, where each variable necessarily takes value in [0, 1]. ut
Furthermore – under an assumption of boundedness – we prove Corollary 1, showing that an MNE alwaysexists.
Proof (Proof of Corollary 1). Let Fi be the feasible region of the i-th player (leader), namely a bounded set.Given x−i, the objective of its optimization problem is linear. Hence, an optimal solution always exists, whichis an extreme point of conv(Fi). However, given that Fi are feasible sets of bilevel linear programs, we knowthat the feasible region of the leaders is a finite union of polyhedra from Theorem 2. It follows that conv(Fi)is a polyhedron. Since we also assume boundedness, conv(Fi) is indeed a polytope. Thus, the i-th player’sstrategy is the set of extreme points of this polytope, finite in number. Since the same reasoning holds foreach player, this is a Nash game with finitely many strategies. From Nash (1950, 1951), such a game has anMNE . ut
From Corollary 1, deciding on the existence of an MNE is trivial if each player has a bounded feasible set.We extend this result with Theorem 5, showing that even if one player’s feasible region is unbounded, thendeciding on the existence of an MNE is Σp
2 -hard.Before proving Theorem 5, we introduce the technical Lemma 1. While Theorem 2 shows that any finite
union of polyhedra can be written as a feasible region of a bilevel problem in a lifted space, Lemma 1 explicitlydescribes this set for a given union of two polyhedra.
Lemma 1. Consider the set S defined as the union of two polyhedra, namely
S =
(h, y, x) ∈ R3+ : h = x; y = 1
∪
(h, y, x) ∈ R3+ : h = 0; y = 0
(4)
S has an extended formulation as a feasible set of a simple bilevel program.
From Lemma 1 we can further derive Lemma 2.
Lemma 2. Suppose S ⊆ Rn1 and T ⊆ Rn2 have an extended formulation as bilevel programs. So does S × T .
Therefore, with Lemmata 1 and 2, we can then prove Theorem 5. Both the proofs for these two lemmas canbe found in the electronic companion.
Proof. Proof of Theorem 5. We reduce SUBSET SUM INTERVAL into a problem of deciding the existence of anMNE for a trivial NASP . Let Q =
∑ki=1 qi. Also, as of Theorem 4, let the Latin game and the Greek game
have Latin and Greek terms, respectively.
10
Latin Game. The Latin game is a Stackelberg game. The variables of the leader and the follower are denotedby Latin alphabets x and y, respectively.
maxx0,...,xk+3r+1
∈Ry0,...,yk∈R
:x0
2+
k∑i=1
qixi + 2(Q+ 1)ξr+1xk+3r+1
− (Q+ 1)
(r∑i=1
2i−1xk+i + pxk+3r+1
)(5a)
subject to xi ≥ 0 ∀ i = 0, . . . , k (5b)
yi ≥ 0 ∀ i = 0, . . . , k (5c)
xi ≥ 1 ∀ i = 0, . . . , k (5d)
xk+3r+1 = xk+2r+i ∀ i = 1, . . . , r (5e)
xk+3r+1 = p+
r∑i=1
2i−1xk+r+i (5f)
x0
2+
k∑i=1
qixi ≤ xk+3r+1 (5g)
(xk+i, xk+r+i, xk+2r+i) ∈ S (as in (4)) ∀ i = 1, . . . , r (5h)
(y0, . . . , yk) ∈ arg miny
k∑i=0
yi :yi ≥ −xiyi ≥ xi − 1
∀ i = 0, . . . , k
(5i)
Greek Game. Similarly, the Greek game is a Stackelberg game, where leader and the follower variables aredenoted by Greek alphabets ξ and χ, respectively.
maxξ0,...,ξr+1
∈Rχ1,...,χr∈R
: (1− x0)ξ0 (5j)
subject to ξi ≥ 0 ∀ i = 1, . . . , r (5k)
χi ≥ 0 ∀ i = 1, . . . , r (5l)
ξi ≤ 1 ∀ i = 1, . . . , r (5m)
p+
r∑i=1
2i−1ξi = ξr+1 (5n)
(χ1, . . . , χr) ∈ arg minχ
r∑i=1
χi :χi ≥ −ξiχi ≥ ξi − 1
∀ i = 0, . . . , r
(5o)
Claim 6 The game defined in (5) is a trivial NASP.
Claim 7 The region of space for x – defined by (5c) and (5i) – is the Cartesian product of (xi : xi ≤0 ∪ xi : xi ≥ 0) for i = 0, . . . , k. Similarly the region of the space for ξ – defined by (5l) and (5o) – is theCartesian product of (ξi : ξi ≤ 0 ∪ ξi : ξi ≥ 0) for i = 1, . . . , k.
The proof of this claim is analogous to the ones of Claims 1 and 2.
Claim 8 xk+3r+1 takes integer values only.
11
Proof (Proof of Claim 8). From (5h), each xk+r+i for i = 1, . . . , r can take a value of either 0 or 1, dependingupon which of the two polyhedra (in the definition of S) the variable falls in. Moreover, since in (5f) the RHSis a sum of integers, the LHS xk+3r+1 is also an integer.
Claim 9 (xk+3r+1)2 =∑ri=1 2i−1xk+i + pxk+3r+1 holds for the Latin game’s feasible set.
Proof (Proof of Claim 9). Consider the set S defined in (4). For a point h = x and y = 1 in the first polyhedra,one can write h = xy. Similarly, for a point h = 0 and y = 0 in the second polyhedron, then h = xy. Thus,the nonlinear equation h = xy is valid for the set S. By multiplying both sides of (5f) with xk+3r+1, one gets
(xk+3r+1)2 = pxk+2r+1 +
r∑i=1
2i−1xk+r+ixk+3r+1
= pxk+3r+1 +
r∑i=1
2i−1xk+r+ixk+2r+i
= pxk+3r+1 +
r∑i=1
2i−1xk+i
The second equality follows from (5e), and the third equality from the fact that h = xy is valid for S and(5h).
Claim 10 Given some ξr+1 ∈ Z between p and t− 1, the Latin player has a profitable unilateral deviationfor any feasible strategy with xk+3r+1 6= ξr+1.
Proof (Proof of Claim 10). Note that if ξr+1 is between p and t− 1, then xk+3r+1 = ξr+1 is feasible for theLatin game. Observe the last two terms of the objective function. From Claim 9, we can rewrite them as(Q+ 1)(2ξr+1xk+3r+1 − x2
k+3r+1). By focusing just on the last two terms, these reach a maximum value forthe feasible choice of xk+3r+1 = ξr+1. We can now argue that the player can never be optimal by choosingxk+3r+1 6= ξr+1. As established in Claim 8, xk+3r+1 is restricted to take integer values, and for any otherchoice xk+3r+1, the deficit in objective value is at least Q+ 1. However, even if each of the other terms taketheir maximum possible value, the largest value they can add to is 0.5 +Q < Q+ 1. the claim follows.
Claim 11 If SUBSET SUM INTERVAL has decision YES, then (2) has a PNE (and hence an MNE).
Proof (Proof of Claim 11). Let s be an integer such that p ≤ s < t and ∀I ⊆ 1, . . . , k,∑i∈I qi 6= s, and let
b1, . . . , br ∈ 0, 1 be the unique r-bit binary representation of s− p. Consider the following pure strategiesfor the players:
xk+3r+1 = s (6a)
xk+2r+i = s i = 1, . . . , r (6b)
xk+r+i = bi i = 1, . . . , r (6c)
xk+i = bis i = 1, . . . , r (6d)
x0 = 1 (6e)
ξ0 = 0 (6f)
ξi = bi i = 1, . . . , r (6g)
ξr+1 = s (6h)
Finally, choose xi ∈ 0, 1 for i = 1, . . . , k such that∑ki=1 qixi is the largest value not exceeding s. Since
it is a YES instance of SUBSET SUM INTERVAL,∑ki=1 qixi ≤ s− 1, and thus the strategy is indeed feasible
for both the players. The Latin player has no feasible profitable deviation. This follows from the fact thatxk+3r+1 cannot be chosen differently due to Claim 10. Moreover, the first two terms in the above strategyalready take the largest possible value not violating (5g). Thus the Latin player has no profitable deviation.Now for the Greek player, since x0 = 1, the objective value is always zero, and cannot be improved. Thus,the strategy in (6) is indeed a PNE .
12
Algorithm 1 Enumeration algorithm to obtain an MNE for a NASP
Input: A description of NASP N = (P 1, . . . , Pn).Output: For each i = 1, . . . , n, xij is a pure-strategy played with probability pij , presenting a mixed-strategy with
support size ki.1: for i = 1, ..., n do2: Enumerate the polyhedra whose union defines the feasible set Fi of P i.3: Fi ← cl conv Fi by applying Theorem 3.4: P i ← objective function of P i and a feasible set of Fi.5: end for6: Solve the facile Nash game N = (P 1, ..., Pn) to obtain either a PNE , (x1, . . . , xn) or show that no PNE exists.
7: if no PNE exists for N then8: There is no MNE for N ; exit returning failure.9: end if
10: for i = 1, ..., n do11: if xi ∈ Fi then12: xi1 ← xi; pi1 ← 1; ki ← 1.13: else14: xi =
∑ki
j=1 ηj xij for xi1, . . . , x
iki ∈ Fi with ηj ≥ 0 and
∑ki
j=1 ηj = 1.
15: pij ← ηj for j = 1, . . . , ki.16: end if17: end for18: return (xij , p
ij) for each i = 1, . . . , n and j = 1, ..., ki.
Claim 12 If SUBSET SUM INTERVAL has decision NO, then (2) has no MNE.
Proof (Proof of Claim 12). Recall xk+3r+1 is forced to be an integer between p and t − 1. For any choiceof xk+3r+1, x0 = 0 is selected and x1, . . . , xk are so that (5g) holds with equality. There is no incentive tochoose x0 = 1, which will contribute to only 0.5 in the objective. However, with x0 = 0, the Greek player canchoose arbitrarily large values of ξ0. Hence, there is always a larger choice of ξ0 which constitute a profitabledeviation. Thus, no equilibrium exists for the game. ut
4 An enumeration algorithm to find MNEs for NASPs
First, we introduce Algorithm 1, which enumerates the polyhedra whose union corresponds to each player’sfeasible region. Then, it finds a pure-strategy Nash equilibrium in the convex hull of each player’s feasibleregions. We prove the equivalence between finding a PNE over the convex hull and the original problem.
The feasible region. Consider the feasible region of a simple Stackelberg game, given by A′u+B′v ≤ b, v ∈SOL(P (u)). Using the KKT conditions of the players in P (u), we can rewrite the latter by an extendedformulation, as
S =
x :Ax ≤ b
z = Mx+ q0 ≤ xi ⊥ zi ≥ 0, ∀ i ∈ C
. (7)
Note that the set S is a union of polyhedra.
Preliminary Enumeration Algorithm. We present Algorithm 1, which exploits the knowledge that eachplayer’s feasible region is a finite union of polyhedra (Basu et al. 2019). Step 2 explicitly enumerates all suchpolyhedra, while Step 3 computes the closure of their convex hull using Theorem 3. Since this convex hull is
13
(a) The players’ feasible regions. From Theorem 2, these arefinite unions of polyhedra. Step 2 of Algorithm 1
(b) With Theorem 3, we compute the convex hull of eachplayer’s feasible region. Step 3 of Algorithm 1
(c) Given the convex hulls, the problem reduces to a MIP(LCP) as of Theorem 1. Step 6 of Algorithm 1
!
!
(d) The solution ? can be interpreted as a convexcombination of feasible strategies. Steps 12 and 14 of
Algorithm 1
Fig. 2: A pictorial reprsentation of Algorithm 1.
also a polyhedron, the game N (defined in Step 6) is a facile Nash game, and we can get a PNE for the gameusing Theorem 1.
Let x be a PNE of N and xi be the strategy of the i-th player. If xi belongs to Fi, then at equilibrium iplays xi in N . If xi does not belong to Fi, it is still contained in cl conv Fi. Thereby, xi can be expressed asa convex combination of points – or strategies – in Fi or a limit of such points. Player i would then playa mixed-strategy where each weight in the convex combination – or δ of Theorem 3 – is the probability ofplaying the corresponding pure-strategy, as in Step 14 of Algorithm 1. We remark that the LCP solved inStep 6 is implemented as a feasibility problem and solved as a MIP . If the user is interested in a specificequilibrium, an objective function can be added to this problem. This gives the user the ability to effectivelyperform equilibria selection, if more than one exists. A visualization of the rationale behind the algorithm isin Figure 2. We formalize the correctness and finite termination of the above procedure in Theorem 7.
Theorem 7. Algorithm 1 terminates finitely and (i) if it returns xij , pij for each i = 1, . . . , n, and j = 1, . . . , ki,
then the strategy profile is indeed an MNE for the NASP, (ii) if it returns failure, then N has no MNE.
Proof (Proof of Theorem 7). For the purpose of this proof, we adopt the same symbols introduced inAlgorithm 1. First, the algorithm terminates in a finite number of steps: all loops in Algorithm 1 are finiteloops, Step 2 ends finitely since there are only finitely many polyhedra (see Theorem 2), and Step 3 is also afinite procedure.
Proof of Statement (i) . Observe that if Algorithm 1 does not return failure, then Step 6 finds PNE x for N .Each player’s objective function is linear, and the distribution for the MNE has finite support. Therefore,one can observe that - for each player i - the following holds:
E((ci + Cix−i
)Txi)
=∑j′
ki∑j=1
p−ij′ pij
(ci + Cix−ij′
)Txij =
(ci + Cix−i
)Txi. (8)
Assume a generic player i has an unilateral profitable deviation †xij , and †pij for i = 1, . . . , `i from xi in
their P i problem. Such a deviation is also a mixed-strategy profile. Consider now the pure-strategy for N
given by∑`i
j=1(†pij † xij). It is feasible for the facile game P i. Therefore, leveraging on the linearity of each
player’s objective function, we can show that this is also a profitable deviation for P i in N , and hence find acontradiction.
14
(ci + Cix−i
)Txi =
∑j′
ki∑j=1
p−ij′ pij
(ci + Cix−ij′
)T (xij)
(9)
≥∑j′
`i∑j=1
p−ij′ † pij
(ci + Cix−ij′
)T (†xij)
(10)
=
ci + Ci
∑j′
p−ij′ x−ij′
T `i∑j=1
†pij † xij
(11)
=(ci + Cix−i
)T `i∑j=1
†pij † xij
(12)
Here (12) follows by plugging the profitable deviation into (8), and exploiting its linearity. Since we havea profitable deviation for the mixed strategy for N , there exists a unilateral deviation for N from x. Thiscontradicts the fact that x is a PNE for N . Therefore, such a deviation cannot exist.
Proof of Statement (ii) . To prove this statement, we prove its contrapositive. Namely, we show that if
N has an MNE , then Step 6 obtains a PNE for N and will not return failure. Therefore, it is sufficientto show that N has a PNE . Let the MNE of N be given by each player i ∈ [n] playing xi1, . . . , x
iki
with
probability pi1, . . . , piki
, respectively. Let xi =∑kij=1 p
ijxij be the a feasible pure-strategy for player i. It follows
that (x1, . . . , xn) is a feasible pure-strategy for N , and we now show it is indeed a PNE for N . Given theabove MNE for N , we know that
∑j′
ki∑j=1
p−ij′ pij(C
ix−ij′ + ci)Txij ≤∑j′
p−ij′ (Cix−ij′ + ci)Txi,∀ xi ∈ Fi.
Due to the linearity of the objective function, it follows that:(Cix−i + ci
)Txi ≤
(Cix−i + ci
)Txi ∀xi ∈ Fi. (13)
If (13) holds for all xi ∈ cl conv(Fi), for all i, then x is a PNE of N and the proof will be complete. First,
we show that (13) holds for xi ∈ conv(Fi). Let xi =∑`j=1 λjx
ij , where xij ∈ Fi and λj ≥ 0 and
∑`j=1 λj = 1.
Now consider the ` inequalities of (13), each one for xij for j = 1, . . . , l. Multiply these inequalities bynon-negative λj on both sides, and add to obtain
(Cix−i + ci
)Txi ≤
∑j=1
λj(Cix−i + ci
)Txij
=(Cix−i + ci
)Txi
In the second instance, to show the same holds for xi ∈ cl conv(Fi), consider a convergent sequence xi1, xi2, . . .
with each xij ∈ conv(Fi) and limj→∞ xij = xi.(Cix−i + ci
)Txi ≤
(Cix−i + ci
)Txij ∀ j = 1, 2, . . .
=⇒ limj→∞
(Cix−i + ci
)Txi ≤ lim
j→∞
(Cix−i + ci
)Txij
=⇒(Cix−i + ci
)Txi ≤
(Cix−i + ci
)T (limj→∞
xij
)
15
=(Cix−i + ci
)Txi
Thus, (13) holds for all xi ∈ cl conv(Fi), and x is indeed a PNE of N . ut
Remark 1. Within the proof of Theorem 7, we never exploit any specific properties of simple Stackelberggames. The only assumption we leverage is that the problem is a linear Nash game (i.e., the objective ofeach player is of the form (ci + Cix−i)Tx). In this case, it is sufficient to solve the problem for PNE in theconvex hull of each player’s feasible set to compute an MNE for the original problem. In this spirit, if one cancompute the convex hull of the player’s feasible region, and if objectives are linear, then every game is aconvex game.
5 Enhancing the algorithm
In this section, we present two enhancements of Algorithm 1. In Section 5.1 we introduce an iterative procedureto approximate the closure of the convex hull of each player feasible set. Thereby, we avoid the possibly costlyand unnecessary enumeration of all the polyhedra defining the feasible sets. In Section 5.2, we tailor thealgorithms to specifically retrieve PNE s as opposed to general MNE s.
5.1 Inner approximation algorithm
While Algorithm 1 is guaranteed to terminate and solve the problem, we introduce a procedure that canimprove computational tractability. The feasible region of a simple Stackelberg game is a finite union ofpolyhedra (see Theorem 2), and their convex hull can be computed using Theorem 3. However, since there maybe exponentially many polyhedra, the convex hull description could become untractably large. Algorithm 1intensively leverage on the complete enumeration of such polyhedra in Step 2. The central intuition is tolimit the enumeration by iteratively refining the convex hull’s description for each player. This procedure isalso valid for an individual Stackelberg game or a bilevel program. However, its importance is more relevantwhen dealing with NASPs, where the computation of this convex hull is essential. The key components ofthis approach are the polyhedral relaxation of the set S defined in (7), and the concept of selected polyhedron.
Definition 10 (Polyhedral relaxation). The polyhedral relaxation of the set S defined in (7) is given bythe set O0 = x : Ax ≤ b, z = Mx+ q, xi ≥ 0, zi ≥ 0 ∀ i ∈ C
Clearly, this set contains cl conv(S), and is hence a relaxation. Also, while S is generally not a polyhedron,its polyhedral relaxation is.
Definition 11 (Selected polyhedron). Let b ∈ 0, 1|C| and let C = c1, . . . , ck. Then, the selectedpolyhedron corresponding to b is P(b) = xci ≤ 0, ∀ i ∈ i : bi = 0
⋂[Mx+ q]ci ≤ 0, ∀ i ∈ i : bi = 1.
We can then formally define the concept of inner approximation.
Definition 12 (Inner Approximation). Let J = j1, . . . , j` ⊆ 0, 1mf . Then the inner approximationdefined by J is IJ = cl conv
(⋃b∈J P(b) ∩ O0
).
Remark 2. The size of the extended formulation of IJ is bounded by O(|J |). To ensure a perfect description,we need a choice of J = 0, 1|C|. However, |J | = 2|C| and a description of cl conv(S) will be exponentiallylarge. Unless P = NP, there cannot be any asymptotical improvements (Bard 1991).
Algorithm 2 presents the inner approximation algorithm – an enhancement to Algorithm 1– to retrievean MNE for NASPs. It iteratively constructs an increasingly accurate inner approximation for the players’feasible regions in the NASP , and seeks for a PNE for this restricted game N (Step 4).
Let F1, . . . ,Fn, be the inner approximations of the feasible sets of player 1, . . . , n. One can compute theconvex hull’s closure for each approximation and solve the associated facile Nash game N . If x is a Nashequilibrium of N , the algorithm checks if x – or the associated mixed-strategy implied by x (similarly to
16
Algorithm 2 Inner approximation to obtain an MNE for a NASP
Input: A description of NASP N = (P 1, . . . , Pn) and J = (J1, ..., Jn) where J i ⊆ 0, 1|Ci| where Ci is the set ofindices of complementarity (⊥) conditions for the i-th player.
Output: For each i = 1, . . . , n, xij is a pure-strategy played with probability pij , presenting a mixed-strategy withsupport size ki.
1: function IterInnerApproxNash(N, J)
2: Fi ←inner approximation defined by J i and Fi ← cl conv Fi .3: P i ← objective function of P i and a feasible set Fi.4: Solve the facile Nash game N = (P 1, ..., Pn) to obtain solution x. . Might fail5: x1, ..., xn ← getDeviation(P, x)6: if xi = NULL for all i = 1, ..., n then7: return x.8: end if9: for i = 1, ...n do
10: if xi 6= NULL then11: bi ←binary encoding of a polyhedron containing xi. J
i ← J i ∪ bi.12: end if13: end for14: return InnerApproxNash(N, J).15: end function
Step 14 of Algorithm 1) – is a Nash equilibrium for the original game N . If this is the case, then the algorithmterminates and returns the equilibrium. Conversely, if this mixed-strategy is not an MNE of N , there exists
a profitable deviation xi for some players such that xi 6∈ Fi. Thereby, we refine the inner approximation
of i-th player’s feasible set by adding a polyhedron containing xi. At each iteration of the algorithm, wekeep on adding polyhedra containing the profitable deviations. However, N may not have a PNE in a giveniteration (Step 4). In this case, we gain no additional knowledge about which polyhedra to add to the innerapproximation. Therefore, we arbitrarily add one or more polyhedra to the feasible region of each player inthe problem, keeping the algorithm running. We define as the extension strategy the criteria by which suchpolyhedra are selected.
Broadly speaking, in optimization problems, a point contained in an inner approximation of the feasibleset is feasible for the original problem and provides a primal bound for the original problem. However, this isnot true in the case of a Nash game. In Remark 3 below, we show that the inner approximation game mighthave a Nash equilibrium while the original game does not. Conversely, we also show that the original gamemight have a Nash equilibrium while an inner approximation does not.
Remark 3 (Inner approximation N might have an MNE but N might not). There might be cases where theinner approximation has no MNE , but the original NASP does. Consider the following players’ problems andtheir inner approximation.
Latin Player: minxξx : x ∈ R, x ≥ 0 (14a)
Greek Player: minξ,χxξ : ξ ∈ [−5, 5];χ ≥ 0;
χ ∈ arg minχχ : χ ≥ ξ − 1;χ ≥ −ξ − 1 (14b)
Using KKT conditions on the follower’s problem, the Greek ’s problem can be rewritten as
minξ,χ,µ
xξ : ξ ∈ [−5, 5];µ1 + µ2 = 1;χ ≥ 0;
0 ≤ µ1 ⊥ χ− ξ + 1 ≥ 00 ≤ µ2 ⊥ χ+ ξ + 1 ≥ 0
The polyhedra P (b) corresponding to b = (0, 0), and b = (1, 1) are empty. The remaining two polyhedra canbe projected to the ξ space as [−5,−1] ∪ [1, 5]. We claim that the problem in (14) has no Nash equilibrium.
17
This is because, irrespective of the Latin player’s decision, an optimal decision for the Greek player is ξ = −5.For such a value of ξ, the Latin player has an unbounded objective. Consider the inner approximation due tothe choice J = (0, 1). The equivalent feasible regions can be expressed as follow.
Latin Player: min ξx : x ∈ R, x ≥ 0 (15a)
Greek Player: min xξ : ξ ∈ R, ξ ∈ [1, 5] (15b)
In (15), the inner approximation is exact for the Latin player and is a strict inner approximation for theGreek player. However, (15) has a PNE (ξ, x) = (0, 1).
Conversely, it can also happen that the original NASP has no MNE , but the inner approximation does.For such an example, replace the objective of the Greek player in (14) with a minimization of −xξ, and thecorresponding inner approximation of the Greek player in (15) with ξ ∈ [−5,−1]. This inner approximationgame has no Nash equilibrium. However, the original game has a Nash equilibrium of (ξ, x) = (0, 5).
5.2 Enhancements for PNEs
In specific applications, deterministic strategies are preferred over randomized ones. Thus, one necessarilyrequires a PNE or show that no PNE exists. With this motivation, we alter Algorithm 1 to retrieve PNE sspecifically or prove no PNE exists.
Enumeration for PNE. This algorithm is similar to Algorithm 1, hence we assume the same notation. First,the procedure explicitly enumerates all the polyhedra in the feasible region of each player, and computes theirconvex hull. In addition, it introduces in N a set of binary variables forcing the equilibrium strategy, for eachplayer, to be strictly in the original feasible region rather than solely in the convex hull. From Theorem 2,the feasible region for each NASP ’s player is a finite union of polyhedra. Let the feasible region of the i-thleader be Fi =
⋃gij=1 P
ij , where P ij = Aijx ≤ bij is a polyhedron. Moreover, Theorem 3 gives cl conv(Fi)
as Aijxij ≤ bijδ
ij for j ∈ [gi], x
i =∑gij=1 x
ij , and
∑gij=1 δ
ij = 1. If for some j, δij = 1, then the projection x is
strictly in the polyhedron P ij . Since we can reformulate a NASP as a MIP feasibility problem, we enforce a
new set of constraints in N requiring each δij to be binary in N . Hence, each PNE for N is also a PNE for
N , and if N has no PNE , also N has no PNE . In addition, for the equivalence between PNE s in N and N ,the condition of N being a NASP can be relaxed. In particular, it is sufficient that leaders’ objectives in Nare convex – observe that under this case, the reasoning in the proof of statement (ii) for Theorem 7 directlyfollows.
We refer the reader to the electronic companion for the pseudocode of this procedure.
6 Computational Tests
We test our algorithms1 with the energy-trade model (18) and (19).
The Model. We consider different geographical regions, where Governments of such regions act as leaders.Governments determine the amount of energy export/import and the CO2 taxation scheme imposed on theirrespective followers (energy producers). Each country seeks to minimize the sum of three components: (i) theproduct between each follower’ production and the emission cost (e.g., the social cost of carbon, SCC), (ii) theproduct between import price and quantity to any other country, (iii) the negative product between importprice and quantity of any other country, namely the maximization of export revenues. Besides, countriesmay also include a negative (maximized) tax-revenue term in their objectives, namely the sum of all theirrespective followers’ taxes. We distinguish between three forms of taxation: (i) Standard-Taxation, whereeach follower has a possibly different tax per unit-energy produced, (ii) Single-Taxation, where every followerhas the same tax per unit-energy produced, (iii) Carbon-Taxation, where every follower has the same taxper unit-emission. The lower-level players are energy producers deciding the amount of production of theirplants based on their linear and quadratic unit costs and their leader’s taxation levels. In specific, followersare playing a Cournot game where the homogeneous good is the amount of energy produced.
18
Instances. We propose three sets of computational instances (InstanceSet A, B, and Insights), and a casestudy on a real-world inspired Chile-Argentina instance 2. The goal of our computational tests is twofold.On the one hand, we showcase our algorithms’ compelling computational capabilities and compare theirperformances. With this respect, we extensively test all our algorithms on the instance sets A and B. On theother hand, we derive managerial insights from our models’ solutions, focusing on the Chile-Argentina casestudy and the instance set Insights.
Data generation. We synthetically generate our instance sets as follow: (i) InstanceSet A contains 150instances with 3 to 5 countries and up to 3 followers per country, (ii) InstanceSet B contains 50 instanceswith 7 countries and up to 3 followers per country, (iii) InstanceSet Insights contains 50 instances with 2countries and 3 followers per country.
We randomly draw each of the followers from three classes of producers: highly-polluting (e.g., coal, oil),averagely-polluting (e.g., gas), and green (e.g., renewables such as solar, hydro). Their emission costs perunit-energy (e.g., GWh) takes an integer value in the range [300, 500], [100, 200], and [25, 50], respectively.These are USD values of emission assuming a social cost of carbon at USD 25 per tonne of CO2 equivalentand typical emission values in these technologies. We set linear and quadratic production costs – negativelycorrelated to the emission factors – in the respective ranges [150, 300], and [0, 0.6] for unit energy. Theproduction capacities are discrete unit-energies in the interval [50, 20000]. We refer the reader to Section 11for a more detailed review of the parameters.
6.1 Strategic insights
Starting from InstanceSet Insights, we solve each instance 4 times by testing a discrete grid of 2 parameters.The first one is the Carbon-Taxation revenue in every country’s objective, while the second dictates whethertrade among countries is allowed. Table 6 provides comprehensive results. We attempt to answer the followingstrategic questions:
(i) Tax policy. Are countries reducing further their emissions if they consider the carbon tax as a source ofincome?
(ii) Trade policy. How does competitive energy trade among countries affect global emission?
Tax policy. Some literature argues that carbon tax revenues can further help reduce carbon emission, spurgreener technologies (e.g., carbon sequestration, electric vehicles) or even to meet other governmental expenses(Olsen et al. 2018, Liu and Lu 2015, Amdur et al. 2014). One might instinctively think that an income-hungry(e.g., GDP) government could levy a more aggressive carbon tax policy if that could be a revenue sourceand help reduce emissions. However, we observe the opposite to be true. We consistently find that when thegovernment’s objective (b = 1 in (18)) model incomes through a carbon tax, the government is systematicallyincentivized to impose a smaller tax. This is because, with smaller tax rates, the production through coal andnatural gas is larger. Thus, this increases the governmental revenue, which is the product of production andtax per unit of emission. In summary, decreased carbon tax could give increased revenue for the government.However, emissions are decreased compared to the no-taxation scheme but increased compared to the casewhen the government does not look for revenue from these taxes.
In particular, in 40 out of the 50 test instances, both countries’ total emission was greater if the individualgovernments considered the objective’s tax revenue. On an absolute basis, emissions were about 13.5% moreon average when governments imposed taxes, keeping the revenue in their objective. A statistical t-test rejectsthe null hypothesis that the global emissions are equal with and without the countries considering carbon taxas a revenue source with a p-value of 0.00018.
In a similar vein, we observe that the trade is lesser in 30 out of 50 instances and, on average, about 7.8%lesser when the countries consider tax as a revenue source. However, a similar t-test does not suggest enoughevidence to reject the null hypothesis (p-value = 0.29) that the traded quantities in the two cases have thesame population mean.
19
Trade policy. Secondly, we observe that the tax rate is typically lesser when countries can trade. Quantitatively,we find that the average tax rate without trade is about 12.9% higher than the tax rate when trade existsbetween the countries. However, we also observe that in 63 of the 100 possible cases (50 instances in InstanceSetInsights with 2 countries each), the tax rate is higher if there is trade between them. In other words, tax isslightly higher in many instances when the trade is enabled. Nevertheless, in those instances where the taxrate is lower with trade enabled, the tax rate is significantly lower.
Next, one might wonder if increased emissions might accompany trade between countries. Since energytrade is an economic activity, one can think it could worsen the externality of emission. However, we observethat emissions are consistently less when countries can trade. This is primarily driven by the fact that cleanmeans of energy in a different country could fulfill the demand in a country without forcing domestic producersto produce using non-green means of production. Quantitatively, we compare the average emission by boththe countries when a trade happens between them instead of no-trade being allowed between them (seeTable 6). We observe that the average emission dropped by about 35.9% when trade was enabled. Further,never in those 50 test instances did the emission ever increase after trade was enabled. We also note thatwhen countries can trade, emissions could increase in one country, but the decrease in another country isalways significant enough to ensure that the total emission decreases while keeping the consumption in bothcountries roughly the same.
Final comments. Besides our consistent insights that (i) a tax revenue-hungry government might imposelesser carbon tax than a government inclined to reduce emissions (ii) enabling trade reduces total globalemissions, the answers to the more general questions were predominantly instance dependent. In particular,we observed that opening up trade increased domestic carbon taxes in some cases and decreased them inothers. We observed similar behaviors for trade with revenue-hungry governments. The answers to thesequestions were sensitive to the cost, capacity, and emission factors of production units and the domesticenergy demand of each country. These observations suggest that one has to solve a NASP (or even a morecomplex model) to identify the specific dynamics for a given situation. Furthermore, in NASPs one canalways perform equilibria selection – if multiple MNE s exist – by solving the problem with Algorithm 1 andenforcing the MIP ’s objective to optimize a given criterion.
6.2 Case study – Chile-Argentina energy markets
We implemented the model using actual data (2018-2019) of Chile’s and Argentina’s electricity markets.Electricity trade between these countries started in 2016, with Chile exporting to Argentina a small amount –close to 1558 MWh – of electricity. However, the transfers are expected to increase as both countries signedan energy cooperation agreement in 2019 (both electricity and gas). These efforts have created some debateregarding electricity prices, which may impact one of the Chilean government’s main goals: make electricitymore affordable. Furthermore, both Chile and Argentina have signed the Paris agreement and promised rapiddecarbonization of their energy systems. Chile was the first country in Latin America to implement a carbontax (USD 5/tCO2), followed by Argentina, which defined a carbon tax that became operational in 2019.Given this context, this analysis focuses on determining the impacts of an integrated market where electricitytrade is viable while each country’s government can define internal carbon tax policies.
We model different energy producers in each country. Electricity producers in Chile and Argentina havevarious technologies. We consider hydro, solar, wind, natural gas, and coal technologies in Chile’s case.Historical data shows that Argentina heavily relies on thermal plants fueled by natural gas and on hydroenergy. Technical data for different technologies, obtained from the Chilean Comision Nacional de Energia(National Energy Agency) and the US Energy Information Administration, include fuel consumption, capacityfactors, and variable costs. We model a stake of coal-based production technology only in Chile, and minimalto none in Argentina. We analyzed how the markets react under different renewable sources’ future levels andwith/without limits on energy trade imposed between these two countries.
If no trade is allowed (representing current operations), we calibrated the model to match historical data(2018/2019) for both countries. There is a significantly larger demand in Argentina (129 TWh/y) than inChile (60 TWh/y). Approximately 71% of the generation in Argentina roots in natural gas thermal power
20
plants. Hydro energy fulfills the remaining demand. In Chile’s case, coal and gas power plants have a marketshare of 42%, hydro accounts for 36%, and renewable sources (solar and wind) supply approximately 15% ofthe electric demand.
When trade is allowed among countries and install capacities are not varied (existing capacities in bothcountries), an interesting substitution effect is observed. Imports from Argentina replace conventional meansof production in Chile (coal and gas). The Chilean government curtails fossil-fueled electricity by increasingthe carbon tax faced by such technologies. The opposite effect shows in Argentina, where the governmentlowers the carbon tax to incentivize electricity generation from natural gas technologies. Such an exporthurts the local market. As expected, an increase in exports to Chile yields increased local electricity prices inArgentina, significantly lowering the indigenous consumption levels.
As observed above, with a possibility for energy trade between the countries, our model predicts thatwithout a significant increase in the renewable capacity on either country or without a significant decreasein carbon’s social cost, Argentina’s economy could be highly impacted. Therefore, unless cheap (near-zero)renewable sources produce energy in Chile or Argentina, it is expected that trade among countries will remainlow.
To assess the likelihood of future trade under large deployments of renewable energy, we consider twocases of increased wind and solar capacity in Chile. The two scenarios consider capacity additions of 20 GWand 40 GW, respectively (Amigo et al. 2021).
In these cases, we initially observe that Chile benefits from increased renewable capacity if energy trade isnot allowed. Electricity prices are reduced by 13% when there is an increase of 40 GW, while consumptiongrows by 20%. Interestingly, when energy trade is allowed, we notice that Argentina becomes a net importerof electricity as Chile increases its renewable energy capacity. When 40 GW of renewable capacity is installed,Argentina has net imports of 12 TWh/y. Therefore, Argentina goes from being a net exporter (withoutrenewable capacity installed) to a net importer of electricity. This import is a direct result of the availabilityof cheap energy, which increases the demand.
6.3 Speed analysis
In terms of performance analysis, we focus on InstanceSet A, and B. An instance is marked a solved if ithas an MNE , or an algorithm finds a certificate of inexistence, namely, no MNE exists. The time limit isTL = 1800 seconds. In our implementation, we introduce 3 extension strategies for Algorithm 2: given alexicographic order for each leader’s polyhedra, k of them are added sequentially, reverse-sequentially, orrandomly.
Table 1: Results summary of different algorithmic configurations for InstanceSetA.
Time (s) WinsAlgorithm ES k EQ NO All EQ NO Solved
FE - - 29.08 0.12 120.21 6 82 140/149
Seq 1 6.65 0.35 51.33 3 0 145/149Seq 3 17.76 0.18 55.82 5 0 145/149Seq 5 6.40 0.15 51.08 3 0 145/149
Rev.Seq 1 7.97 0.36 3.73 26 0 149/149Rev.Seq 3 11.29 0.18 53.12 4 0 145/149Rev.Seq 5 9.53 0.15 76.41 5 0 143/149Random 1 5.22 0.36 26.60 8 0 147/149Random 3 32.42 0.18 85.65 5 0 143/149
MNEInnerApp
Random 5 23.67 0.15 58.26 2 0 145/149
PNE FE-P - - 7.25 0.12 328.23 – – 122/149
Tables 1 and 2 summarize the computational results for InstanceSetA and InstanceSetB, respectively.The upper parts of the tables reports results for the full enumeration Algorithm 1 (FE ) and Algorithm 2
21
Table 2: Results summary of different algorithmic configurations for InstanceSetB.
Time (s) WinsAlgorithm ES k EQ NO All EQ NO Solved
FE - - 260.29 1.12 1174.32 0 2 20/50
Seq 1 39.26 9.64 672.24 1 0 32/50Seq 3 62.66 3.88 616.25 1 0 34/50Seq 5 24.03 2.83 733.97 1 0 30/50
Rev.Seq 1 171.47 9.66 262.74 27 0 47/50Rev.Seq 3 13.85 3.86 585.27 4 0 34/50Rev.Seq 5 78.57 2.83 798.90 6 0 29/50Random 1 34.65 9.65 497.06 0 0 37/50Random 3 123.02 3.87 588.03 2 0 36/50
MNEInnerApp
Random 5 39.18 2.86 711.77 4 0 41/50
PNE FE-P - - 7.36 1.12 1441.95 – – 10/50
(InnerApp), where an MNE solves the instances. In the bottom part, we specifically look for PNEs withthe enhanced algorithm presented in Section 5.2 (FE-P). In the third column, if the algorithm is the innerapproximation, we highlight the extension strategies, and the relative parameter k in the following column.Fifth, sixth, and seventh columns are, respectively, average time when: (i) an MNE is found (EQ), (ii) acertificate of non-existence is returned (NO) and (iii) for all instances. In the eighth and ninth column, wereport the number of times the row’s algorithm outperforms all the others, namely wins in terms of computingtimes. Finally, the tenth column reports how many instances do not trigger the time limit.
For MNE s, InnerApp achieves better performances than FE, being on average 2x faster on all instances,and up to 30x when an MNE exists (see InnerApp-RevSeq-1 in Table 1). Table 2 shows the full potentialof InnerApp, which remarkably reduces computational times compared to FE. Especially, InnerApp cansolve almost all the 50 hard instances compared to the 20 solved by FE. Besides, when no equilibrium exists,InnerApp will always terminate at its last iteration, namely the one corresponding to FE. It is not surprisingthat FE returns a non-existence certificate always faster than InnerApp. Both the algorithms InnerApp andFE – when asked to retrieve a generic MNE – may return a PNE . This happens 37.6%, and 30.4% withinInstanceSetA and InstanceSetB, respectively. Hence, there is a natural need for FE-P.
7 Concluding Remarks
Our theoretical and computational framework tackles NASPs, where players of a Nash game solve linearbilevel programs, and each leader can have several followers playing a simple Nash game among themselves.We have shown that deciding existence of PNE and MNE for NASPs is Σp
2 -hard, and we provided a familyof algorithms to find MNE s as well as PNE s for the problem. We shown it is sufficient to compute an MNEover the convex hull of each player’s feasible region to retrieve a MNE for the original problem.
This work expands our knowledge of algorithmic approaches to compute equilibria, in particular MNE s.In addition to a theoretical characterization of these algorithmic methods, it analyzes their practical efficiency,settles their limitations, and opens up new future directions by establishing a solid benchmark againstwhich future progress can be measured. From an application standpoint, we demonstrated how the NASPsframework could help unveil counterintuitive consequences of policymaking within the context of internationalenergy trade.
In terms of forthcoming work, the computation of multiple equilibria, or their selection according to somespecified criteria, is interesting. Furthermore, it may be worth developing procedures to prune parts of thefeasible regions (e.g., polyhedra) not in the support of any equilibrium. This last direction would remarkablyspeed up equilibria computation. Any advancements on these proposed research lines would enable us totackle more general cases, for instance, where interactions among followers of different leaders are allowed.
22
Endnotes
1. Full implementation with detailed documentation are available on https://github.com/ssriram1992/
EPECsolve.2. All instances are available on https://github.com/ds4dm/EPECInstances.
23
Bibliography
Amdur D, Rabe BG, Borick CP (2014) Public views on a carbon tax depend on the proposed use of revenue. Issues inEnergy and Environmental Policy -(13).
Amigo P, Cea-Echenique S, Feijoo F (2021) A two stage cap-and-trade model with allowance re-trading and capacityinvestment: The case of the chilean ndc targets. Energy 120129.
Balas E (1985) Disjunctive Programming and a Hierarchy of Relaxations for Discrete Optimization Problems. SIAMJournal on Algebraic Discrete Methods 6(3):466–486, ISSN 0196-5212, URL http://dx.doi.org/10.1137/
0606047.
Bard JF (1991) Some properties of the bilevel programming problem. Journal of Optimization Theory and Applications68(2):371–378, ISSN 1573-2878, URL http://dx.doi.org/10.1007/BF00941574.
Bard JF, Plummer J, Sourie JC (1998) Determining tax credits for converting nonfood crops to biofuels: An applicationof bilevel programming. Migdalas A, Pardalos PM, Varbrand P, eds., Multilevel Optimization: Algorithms andApplications, 23–50 (Boston, MA: Springer US), ISBN 978-1-4613-0307-7.
Bard JF, Plummer J, Sourie JC (2000) A bilevel programming approach to determining tax credits for biofuelproduction1this work was supported by a grant from institut national de la recherche agronomique and thetexas higher education coordinating board under the advanced research program – arp 003.1. European Journalof Operational Research 120(1):30 – 46, ISSN 0377-2217, URL http://dx.doi.org/10.1016/S0377-2217(98)
00373-7.
Basu A, Ryan CT, Sankaranarayanan S (2019) Mixed-integer bilevel representability. Mathematical ProgrammingISSN 1436-4646, URL http://dx.doi.org/10.1007/s10107-019-01424-w.
Boffey D (2021a) Eu threatens to block covid vaccine exports amid astrazeneca short-fall. The Financial Times URL https://www.theguardian.com/world/2021/jan/25/
eu-threatens-to-block-covid-vaccine-exports-amid-astrazeneca-shortfall, accessed on Marchthe 4th 2021.
Boffey D (2021b) Italy blocks export of 250,000 astrazeneca vaccine doses to aus-tralia. The Guardian URL https://www.theguardian.com/world/2021/mar/04/
italy-blocks-export-of-250000-astrazeneca-vaccine-doses-to-australia, accessed on March the4th 2021.
Brotcorne L, Labbe M, Marcotte P, Savard G (2008) Joint design and pricing on a network. Operations Research56(5):1104–1115, URL http://dx.doi.org/10.1287/opre.1080.0617.
Candler W, Norto R (1977) Multi-level programming and development policy. Working Paper 1–56.
Candler W, Townsley R (1982) A linear two-level programming problem. Computers & Operations Research 9(1):59–76.
Carvalho M, Lodi A, Pedroso JP (2018a) Existence of nash equilibria on integer programming games. Vaz AIF,Almeida JP, Oliveira JF, Pinto AA, eds., Operational Research, 11–23 (Cham: Springer International Publishing),ISBN 978-3-319-71583-4.
Carvalho M, Lodi A, Pedroso JP (2020) Computing Nash equilibria for integer programming games.
Carvalho M, Lodi A, Pedroso JP, Viana A (2017) Nash equilibria in the two-player kidney exchange game. MathematicalProgramming 161(1):389–417, ISSN 1436-4646, URL http://dx.doi.org/10.1007/s10107-016-1013-7.
Carvalho M, Pedroso JP, Telha C, Vyve MV (2018b) Competitive uncapacitated lot-sizing game. International Journalof Production Economics 204:148 – 159, ISSN 0925-5273, URL http://dx.doi.org/https://doi.org/10.1016/
j.ijpe.2018.07.026.
Cashell B, Jackson WD, Jickling M, Webel B (2004) The economic impact of cyber-attacks. Congressional ResearchService Documents, CRS RL32331 (Washington DC) .
Chen X, Deng X (2006) Settling the complexity of two-player nash equilibrium. 2006 47th Annual IEEE Symposiumon Foundations of Computer Science (FOCS’06), 261–272 (IEEE).
Cottle R, Pang JS, Stone RE (2009) The Linear Complementarity problem (Society for Industrial and AppliedMathematics (SIAM, 3600 Market Street, Floor 6, Philadelphia, PA 19104)), ISBN 898719003.
Debreu G (1952) A Social Equilibrium Existence Theorem. Proceedings of the National Academy of Sciences 38(10):886–893, ISSN 0027-8424, 1091-6490, URL http://dx.doi.org/10.1073/pnas.38.10.886.
Del Pia A, Ferris M, Michini C (2017) Totally unimodular congestion games. Proceedings of the Twenty-Eighth AnnualACM-SIAM Symposium on Discrete Algorithms, 577–588, SODA ’17 (USA: Society for Industrial and AppliedMathematics).
DeMiguel V, Xu H (2009) A stochastic multiple-leader stackelberg model: analysis, computation, and application.Operations Research 57(5):1220–1235.
Eggermont CE, Woeginger GJ (2013) Motion planning with pulley, rope, and baskets. Theory of Computing Systems53(4):569–582.
Egging R, Gabriel SA, Holz F, Zhuang J (2008) A complementarity model for the european natural gas market.Energy policy 36(7):2385–2414.
Egging R, Holz F, Gabriel SA (2010) The world gas model: A multi-period mixed complementarity model for theglobal natural gas market. Energy 35(10):4016–4029.
Facchinei F, Pang JS (2015a) Finite-Dimensional Variational Inequalities and Complementarity Problems, Vol 1,volume 1 (Springer-Verlag), ISBN 9788578110796, URL http://dx.doi.org/10.1017/CBO9781107415324.004.
Facchinei F, Pang JS (2015b) Finite-Dimensional Variational Inequalities and Complementarity Problems, Vol 2,volume 2 (Springer-Verlag), ISBN 9788578110796, URL http://dx.doi.org/10.1017/CBO9781107415324.004.
Fan K (1952) Fixed-point and Minimax Theorems in Locally Convex Topological Linear Spaces. Proceedings of theNational Academy of Sciences 38(2):121–126, ISSN 0027-8424, 1091-6490, URL http://dx.doi.org/10.1073/
pnas.38.2.121.
Feijoo F, Das TK (2014) Design of pareto optimal co2 cap-and-trade policies for deregulated electricity networks.Applied energy 119:371–383.
Feijoo F, Huppmann D, Sakiyama L, Siddiqui S (2016) North american natural gas model: Impact of cross-bordertrade with mexico. Energy 112:1084–1095.
Feijoo F, Iyer GC, Avraam C, Siddiqui SA, Clarke LE, Sankaranarayanan S, Binsted MT, Patel PL, Prates NC,Torres-Alfaro E, et al. (2018) The future of natural gas infrastructure development in the united states. Appliedenergy 228:149–166.
Fleming S, Brunsden J, Miles (2021) Italy blocks shipment of oxford/astrazeneca vaccine to australia. The FinancialTimes URL https://www.ft.com/content/bed655ac-9285-486a-b5ad-b015284798c8, accessed on March the4th 2021.
Gabriel SA, Conejo AJ, Fuller JD, Hobbs BF, Ruiz C (2012) Complementarity Modeling in Energy Markets (Springer-Verlag), ISBN 9781441961235.
Gabriel SA, Leuthold FU (2010) Solving discretely-constrained mpec problems with applications in electric powermarkets. Energy Economics 32(1):3–14.
Gilboa I, Zemel E (1989) Nash and correlated equilibria: Some complexity considerations. Games and EconomicBehavior 1(1):80 – 93, ISSN 0899-8256, URL http://dx.doi.org/https://doi.org/10.1016/0899-8256(89)
90006-7.
Glicksberg IL (1952) A Further Generalization of the Kakutani Fixed Point Theorem, with Application to NashEquilibrium Points. Proceedings of the American Mathematical Society 3(1):170, ISSN 00029939, URL http:
//dx.doi.org/10.2307/2032478.
Gordon LA, Loeb MP (2002) The economics of information security investment. ACM Transactions on Informationand System Security (TISSEC) 5(4):438–457.
Hobbs BF, Metzler CB, Pang JS (2000) Strategic gaming analysis for electric power systems: An mpec approach.IEEE transactions on power systems 15(2):638–645.
Holz F, Von Hirschhausen C, Kemfert C (2008) A strategic model of european gas supply (gasmod). Energy Economics30(3):766–788.
Hu X, Ralph D (2007) Using EPECs to Model Bilevel Games in Restructured Electricity Markets with LocationalPrices. Operations Research 55(5), URL http://dx.doi.org/10.1287/opre.1070.0431.
Jeroslow RG (1985) The polynomial hierarchy and a simple model for competitive analysis. Mathematical programming32(2):146–164.
25
Kulkarni AA, Shanbhag UV (2014) A shared-constraint approach to multi-leader multi-follower games. Set-valued andvariational analysis 22(4):691–720.
Kulkarni AA, Shanbhag UV (2015) An existence result for hierarchical stackelberg v/s stackelberg games. IEEETransactions on Automatic Control 60(12):3379–3384.
Labbe M, Violin A (2013) Bilevel programming and price setting problems. 4OR 11(1):1–30, ISSN 1614-2411, URLhttp://dx.doi.org/10.1007/s10288-012-0213-0.
Leyffer S, Munson T (2010) Solving multi-leader–common-follower games. Optimisation Methods & Software 25(4):601–623.
Li H, Meissner J (2011) Competition under capacitated dynamic lot-sizing with capacity acquisition. InternationalJournal of Production Economics 131(2):535 – 544, ISSN 0925-5273, URL http://dx.doi.org/10.1016/j.ijpe.
2011.01.022.
Liu Y, Lu Y (2015) The economic impact of different carbon tax revenue recycling schemes in china: A model-basedscenario analysis. Applied Energy 141:96–105.
Nash J (1951) Non-Cooperative Games. Annals of Mathematics 54(2):286–295.
Nash JF (1950) Equilibrium Points in N-Person Games. Proceedings of the National Academy of Sciences of the UnitedStates of America 36(1):48–9, ISSN 0027-8424, URL http://dx.doi.org/10.1073/pnas.36.1.48.
Olsen DJ, Dvorkin Y, Fernandez-Blanco R, Ortega-Vazquez MA (2018) Optimal carbon taxes for emissions targets inthe electricity sector. IEEE Transactions on Power Systems 33(6):5892–5901.
Ralph D, Smeers Y (2006) EPECs as models for electricity markets. 2006 IEEE PES Power Systems Conference and Ex-position, 74—-80, URL http://www3.eng.cam.ac.uk/~dr241/Papers/Ralph-Smeers-EPEC-electricity.pdf.
Rosenthal RW (1973) A class of games possessing pure-strategy nash equilibria. International Journal of Game Theory2:65 – 67.
Sagratella S, Schmidt M, Sudermann-Merx N (2019) The noncooperative fixed charge transportation problem. EuropeanJournal of Operational Research ISSN 0377-2217, URL http://dx.doi.org/https://doi.org/10.1016/j.ejor.
2019.12.024.
Sankaranarayanan S, Feijoo F, Siddiqui S (2018) Sensitivity and covariance in stochastic complementarity problems withan application to North American natural gas markets. European Journal of Operational Research 268(1):25–36,ISSN 0377-2217, URL http://dx.doi.org/10.1016/J.EJOR.2017.11.003.
Sherali HD (1984) A multiple leader stackelberg model and analysis. Operations Research 32(2):390–404.
Stackelberg H (1934) Von, 1934, marktform und gleichgewicht. Vienna: Julius Springer English translation availableat https://doi.org/10.1007/978-3-642-12586-7.
Stein O, Sudermann-Merx N (2018) The noncooperative transportation problem and linear generalized nash games.European Journal of Operational Research 266(2):543 – 553, ISSN 0377-2217, URL http://dx.doi.org/10.
1016/j.ejor.2017.10.001.
Weintraub R, Bitton A, Rosenberg M (2020) The danger of vaccine nationalism. Harvard Business Review URLhttps://hbr.org/2020/05/the-danger-of-vaccine-nationalism, accessed on March the 4th 2021.
26
Supplementary Material
In this electronic companion, we complement the proofs of Section 3 in Section 8, the pseudo-code for thePNE s algorithm in Section 10, and an overview of computational instances in Section 11.
8 Extensions to proofs of hardness
Proof (Proof of Claim 1). All the constraints are linear, and if the variables of the other player are fixed, theobjectives are also linear. Also, the follower is simply parameterized in their leader’s variables. There areprecisely two leaders, and their interaction follows the definition of a simple Nash game. Hence – by definition– the game in (2) is a trivial NASP .
Proof (Proof of Claim 2). Notice that the constraints in (2h) enforce yi ≥ max(−xi, xi − 1), and since yi isminimized, it has necessarily to be equal to max(xi−1,−xi). However, if this quantity should be non-negative– as enforced in (2c) – then either xi ≤ 0 or 1− xi ≤ 0 should hold. The claim follows.
Proof (Proof of Lemma 1). The following bilevel problem gives the necessary extended formulation. Variablesz1, z2, . . . are the variables in the lifted space, which can be projected out.
x ≥ 0 (16a)
y ≥ 0 (16b)
h ≥ 0 (16c)
y ≤ 1 (16d)
h ≤ x (16e)
z1, . . . , z6 ≥ 0 (16f)
(z1, , . . . , z6) ∈ arg minz
6∑i=1
zi :
z1 ≥ h− x ; z1 ≥ −hz2 ≥ 1− y ; z2 ≥ −hz3 ≥ y − 1 ; z3 ≥ −hz4 ≥ x− h ; z4 ≥ −yz5 ≥ h− x ; z5 ≥ −yz6 ≥ y − 1 ; z6 ≥ −y
(16g)
Proof (Proof of Lemma 2). If S has an extended formulation given by (x, y) : ASx + BSy ≤ bS ; y ∈arg minfTS y : CSx + DSy ≤ gS, and if T has an extended formulation given by (x, y) : ATx + BT y ≤bT ; y ∈ arg minfTT y : CTx+DT y ≤ gT , then the following is an extended formulation of S × T :
(x, y, u, v) : ASx+BSy ≤ bS ;ATu+BT v ≤ bT ;
(y, v) ∈ arg minfTS y + fTT v :CSx+DSy ≤ gSCTu+DT y ≤ gT
Proof (Proof of Claim 6.). All constraints are linear, and if the variables of the other player are fixed, theobjectives are also linear. The constraints (5h) are valid due to Lemma 1. Also, for Lemma 2, we can havemultiple bilevel constraints in (5h) and (5i). Each follower is simply parameterized in their leader’s variables.There are precisely two leaders, and their interaction follows the definition of a simple Nash game.
27
9 NASP with no PNE but only an MNE
Example 1. Considering the following Latin-Greek trivial NASP .Latin Player
maxx,y
: x1ξ1 + x2ξ2 (17a)
x, y ≥ 0 (17b)
x ≤ 1 (17c)
x1 + x2 = 1 (17d)
y ∈ arg miny
y1 + y2 :
yi ≥ −xiyi ≥ xi − 1
for i = 1, 2
(17e)
Greek Player
maxξ,χ
: x2ξ1 + x1ξ2 (17f)
ξ, χ ≥ 0 (17g)
ξ ≤ 1 (17h)
ξ1 + ξ2 = 1 (17i)
χ ∈ arg minχ
χ1 + χ2 :
χi ≥ −ξiχi ≥ ξi − 1
for i = 1, 2
(17j)
The only feasible decisions for both the Latin and the Greek player in (17) are (1, 0, 0, 0), (0, 1, 0, 0). So thegame can be written as a normal form game. The payoffs for these finitely many strategies can be computedto be that if the Latin and the Greek player choose the same strategy, then the Latin player gets a payoff of 1and the Greek player gets a payoff of 0. If they choose different strategies, then the Latin player gets a payoffof 0, and the Greek player gets a payoff of 1. One can easily check that this game’s unique Nash equilibriumis an MNE and that no PNE exists.
10 Enumeration algorithm to obtain a PNE
Algorithm 3 reports the pseudo-code for the algorithm described in Section 5.2.
Algorithm 3 Enumeration algorithm to obtain a PNE for a NASP
Input: A description of NASP N = (P 1, . . . , Pn).Output: For each i = 1, . . . , n, a pure-strategy xi, such that the strategy profile is a PNE or a proof that no PNE
exists.1: for i = 1, ..., n do2: Enumerate the polyhedra whose union defines the feasible set Fi of P i.3: Fi ← cl conv Fi by applying Theorem 3.4: P i ← objective function of P i and a feasible set of Fi.5: end for6: N = (P 1, ..., Pn) the facile Nash game.
7: Enforce δij for i = 1, . . . , n, j = 1, . . . , gi in N to be binary.
8: if N is infeasible then9: return No PNE exists.
10: else11: return Project the solution of N to the space of the original variables of N .12: end if
28
11 Computations
Governments act as Stackelberg leaders by trading energy, intending to minimize their emissions, andeventually to maximize tax incomes. Within each country, energy producers act as Stackelberg followersand play a Nash game between themselves, aiming to maximize their profits. Each country is interestedin imposing a tax that is not preventing profitable domestic production, as it is constrained to keep thedomestic energy price less than a predetermined threshold. We present the optimization problems of theplayers formally below. For ease of understanding, the quantities in red are parameters, i.e., inputs to themodel. And the quantities in blue are decision variables, decided by the country or of the energy producersin the same country. Quantities in green are variables in a problem, but not decided by the country beingconsidered. Each country C solves the following problem.
minqp,tp,
qC′→Cimp ,qC
exp
:
∑p∈P
Cpemmisionqp − btpqp
+∑
C′∈C\C
πCqC′→C
imp − πCqCexp (18a)
subject to tp ≤ tp (18b)
αC − βC∑p∈P
qp + qCimp − qCexp
≥ πC (18c)
∑C′∈C
qC′→C
imp = qCimp (18d)
qp ∈ SOL(Lower Level Nash Game) (18e)
Cpemmision is the dollar value of the emission caused by producer p while producing a unit quantity of energy.
This number is the product of cost incurred due to the emission of one unit of greenhouse gases (GHG),some times referred to as the social cost of carbon and the quantity of GHG emitted for each unit of energyproduced by the producer p, called as the emission factor. b dictates whether the objective should include thetax revenue earned by the government or not. qp is the quantity of energy produced by the producer p ∈P,qCimp,q
Cexp are respectively import and export quantities, and αC , βC are the intercept and the slope of the
demand curve. The domestic price, for each country, is given by αC − βCQ, where Q is the quantity of energyavailable domestically. Finally, πC is the price at which the country can import energy from other countries,hence the variable linking the optimization problems of different countries. Thus, πC can be interpreted asthe shadow price to the market-clearing constraint∑
C′∈C
qC→C′
imp =∑C∈C
qCexp. (18f)
We note that including (18f) does not make the game into a generalized Nash game. This is because(18f) can be written as the KKT conditions of a fictitious optimization problem, generally referred to as theinvisible hand in the market. An alternative manner of looking at this is as if there is a perfect competitionin the international energy markets and the most efficient allocation of resources happens. This is again astandard simplifying assumption considered, for example, in Egging et al. (2010, 2008), Gabriel and Leuthold(2010), Sankaranarayanan et al. (2018), Feijoo et al. (2018).
Optionally for some countries, as a domestic policy, we introduce a carbon tax paradigm, where thetax imposed on the followers is proportional to the emissions they cause. i.e., there is a constraint tp =Cp
emmisiontGHG, where the government decides the tax payable per unit emission. Furthermore, note that if bis non-zero, the objective is no longer linear. In such a case, we replace the product term with a McCormickrelaxation. Finally, tp, and πC are the tax cap and price cap, respectively. The lower level problem that eachproducer p solves is formulated as follow:
minqp
: Cpqp +
1
2Dpq
p2 + tpqp −
αC − βC∑p′∈P
qp′+ qCimp − qCexp
qp (19a)
29
subject to qp ≥ 0 (19b)
qp ≤ qp (19c)
The first two terms in the objective correspond to the energy producer’s cost, while the third term is the taxexpense. The parenthesis results in the revenue of p, which is the product of domestic price and the quantityproduced. Further, the producer is constrained by their capacity limits. Note that the product of variables(tpqp) in the objective does not pose any additional difficulty to the problem. This is because the follower’sproblem is still convex quadratic for a fixed value of tp, and the KKT conditions give complementarityconstraints with only linear terms.
Further, we also note that the previously-mentioned assumption of optimistic equilibrium selection by theleaders and the limitations imposed as a result, are irrelevant here. This is because, for Dp > 0, which isalways the case, the test examples have a unique lower-level equilibrium. Thus any technique for equilibriumselection is not warrented.
11.1 Instance sets
We generated three instances sets for our computations. (i) InstanceSetA contains 149 instances where there are3 to 5 countries (ii) InstanceSetB contains 50 instances with strictly 7 countries. These instances were selectedif Algorithm 1 was not able to solve them within 10 second on a single core machine. (iii) InstanceSetInsightscontains 50 instances with 2 countries with 3 followers each. Such instances are useful to derive managerialinsights from our model. The specific parameters for all these instances are described in Table 3 and areavailable in our open-source GitHub repository. All our tests run on a 8-cores Intel(R) Xeon Gold 6142, with32GB of RAM and Gurobi 9.0.
Parameter Distribution Notes
Capacities 50, 100, 130, 170, 200, 1000, 1050, 20000 Each follower’s capacity is randomly drawn from these values.
Emission Costs 25, 50, 100, 200, 300, 500, 550, 600 The first two values are reserved for green producers. The following two foraveragely-polluting producers, while the remaining three for highly-pollutingones.
Linear Costs 150, 200, 220, 250, 275, 290, 300 Linear costs are generally inversely proportional to the emission cost. Forinstance, a producer with a 50 emission cost will generally have a linear costaround 290.
Quadratic Costs 0, 0.1, 0.2, 0.3, 0.5, 0.55, 0.6 Quadratic costs are generally inversely proportional to the emission cost.Same rationale as linear costs.
Tax Caps 0, 50, 100, 150, 200, 250, 275, 300 Tax caps are assigned following the same rational of emission costs. Thelower the emission cost of a given producer, the lower the maximum taxapplicable to it.
Demand Alpha 275, 300, 325, 350, 375, 450 Each country alpha is randomly drawn from this set.
Demand Beta 0.5, 0.6, 0.7, 0.75, 0.8, 0.9 Each country beta is randomly drawn from this set.
Price Cap 0.8, 0.85, 0.90, 0.95 Each country price-limit is randomly drawn from this set. The final price-limit is made of the product of this value and the country’s demand alphaparameter.
Tax Paradigm Standard, Single, Carbon A country tax scheme can be: (i.) Standard, where followers are taxed at dif-ferent levels per unit-energy, or (ii. ) Single, where all the followers are taxedwith the same level per unit-energy (iii.) Carbon, where all the followers aretaxed with the same level per unit-emission
Table 3: Description of the parameters for our instances.
.
11.2 Results tables
Tables 4 and 5 contains the full results for InstanceSetA and InstanceSetB, respectively. The first threecolumns are the instance number, the number of leaders, and – for each leader – their respective number
30
of followers in squared parenthesis. The MNE column is the status of the instance, namely if it has anequilibrium (YES ), if it does not (NO), or if the time limit was triggered (TL) for all the methods. In theremaining column, we report the clock time and the status for each algorithmic configuration. In particular,we have Algorithm 1 (FE ), and the inner approximations. We report three extension strategies, namelythe sequential (seq), the reverse sequential (rseq), and the random one (rand). They are followed by theirrespective parameter k, as reported in Section 6. The last two columns are related to PNE s.
Table 6 reports the results for InstanceSet Insights. The first column reports each instance’s number.The second and third column are boolean values reporting whether the tax (Ta) and the trade (Tr) areallowed (value of 1) or not. The following 16 columns are results for the first country (Country One), whilethe remaining 16 are for the second country (Country Two). Following the column order, for each country wereport: the unit-energy production level Prod, the domestic price per unit-energy $(E), the import Imp andexport Exp unit-energies, the export price $(E), and the tax per unit-emission Tax. Furthermore, for each ofthe the 3 followers of each country, we have the type Ty (C for coal, G for gas, or S for solar), the associatedemission cost per unit-energy E, and its production Prod.
31
Table
4:
MN
Eand
PN
Ere
sult
sfo
rIn
stan
ceS
etA
.C
olu
mns:
#-
Inst
ance
Num
ber
.L
-N
um
ber
of
leader
sin
the
inst
ance
.F
-N
um
ber
of
foll
ower
sea
chle
ader
has
.F
E-
Tim
eta
ken
for
full
enu
mer
ati
on
alg
ori
thm
.se
q1to
ran
d5
-T
ime
take
nfo
rin
ner
ap
pro
xim
ati
on
wit
hd
iffer
ent
exte
nsi
on
stra
tegie
s.M
NE
-ex
iste
nce
(or
tim
eli
mit
reach
edT
L).
FE
-P-
Tim
efo
rfu
llen
um
erati
on
tofi
nd
aP
NE
.P
NE
-ex
iste
nce
(or
tim
eli
mit
reac
hed
TL
).
#L
FFE
seq1
seq3
seq5
rseq1
rseq3
rseq5
rand1
rand3
rand5
MNE
FE-P
PNE
13
[1
22
]0.0
40.0
40.0
40.0
40.0
40.0
40.0
40.0
40.0
40.0
4NO
0.0
4NO
23
[2
23
]0.0
70.1
50.0
80.0
80.1
60.0
80.0
80.1
60.0
80.0
8NO
0.0
7YES
33
[2
22
]0.0
50.1
30.0
60.0
60.1
30.0
60.0
60.1
30.0
60.0
6NO
0.0
5NO
43
[1
21
]0.0
40.1
00.0
50.0
50.1
00.0
50.0
50.1
00.0
50.0
5NO
0.0
4TL
53
[2
13
]0.3
30.0
40.0
40.0
40.0
40.0
40.0
40.0
40.0
40.0
4YES
0.0
9YES
63
[2
21
]0.0
60.1
80.1
10.0
70.1
80.1
10.0
70.1
80.1
10.0
7NO
0.0
6YES
73
[2
22
]0.0
60.1
40.0
70.0
70.1
40.0
70.0
70.1
40.0
70.0
7NO
0.0
6YES
83
[1
21
]0.1
00.0
80.0
80.0
80.0
80.0
80.0
80.0
80.0
80.0
8YES
0.0
8NO
93
[1
22
]0.0
50.1
00.0
60.0
60.1
00.0
60.0
60.1
00.0
60.0
6NO
0.0
4NO
10
3[2
21
]0.1
60.2
40.1
70.1
70.1
30.1
90.2
00.2
40.2
00.2
0YES
0.1
6NO
11
3[3
22
]0.0
80.2
20.1
50.0
90.2
30.1
40.0
90.2
20.1
40.0
9NO
0.0
8NO
12
3[2
22
]0.7
61.5
01.4
11.0
70.3
80.2
80.7
91.7
21.7
70.5
5YES
2.0
7YES
13
3[2
12
]0.0
40.0
70.0
50.0
50.0
70.0
50.0
50.0
70.0
50.0
5NO
0.0
4NO
14
3[2
13
]0.0
60.1
30.0
70.0
70.1
30.0
70.0
70.1
30.0
70.0
7NO
0.0
6NO
15
3[2
12
]0.3
90.0
80.0
80.0
80.0
90.0
80.0
80.0
80.0
80.0
8YES
0.1
8NO
16
3[2
32
]5.7
35.0
77.6
84.8
30.3
71.4
812.1
913.4
91370.2
81.3
0YES
TL
YES
17
3[2
22
]0.0
60.1
30.0
70.0
70.1
30.0
70.0
70.1
30.0
70.0
7NO
0.0
6YES
18
3[2
22
]0.7
90.1
50.1
60.1
50.1
50.1
50.1
50.1
50.1
50.1
5YES
3.1
7NO
19
3[2
22
]0.0
70.1
60.0
80.0
80.1
60.0
80.0
80.1
60.0
80.0
8NO
0.0
7NO
20
3[1
32
]0.0
50.0
90.0
60.0
60.0
90.0
60.0
60.0
90.0
60.0
6NO
0.0
5NO
21
3[1
22
]0.0
40.1
10.0
60.0
60.1
20.0
60.0
60.1
10.0
60.0
6NO
0.0
5YES
22
3[2
11
]0.0
40.1
00.0
50.0
50.1
00.0
50.0
50.1
00.0
50.0
5NO
0.0
4NO
23
3[1
21
]0.0
90.1
30.1
10.1
10.1
00.1
10.1
10.1
30.1
10.1
1YES
0.0
7YES
24
3[2
21
]0.0
50.1
40.0
90.0
60.1
40.0
90.0
60.1
40.0
90.0
6NO
0.0
5YES
25
3[2
22
]0.1
40.2
40.1
70.1
70.1
40.1
60.1
70.1
20.1
70.1
7YES
0.1
2NO
26
3[2
22
]0.0
60.1
50.0
70.0
70.1
60.0
70.0
70.1
50.0
80.0
7NO
0.0
6NO
27
3[1
12
]0.1
00.0
40.0
40.0
40.0
40.0
40.0
40.0
40.0
40.0
4YES
0.1
2NO
28
3[3
13
]0.2
61.1
00.5
50.9
60.1
00.3
30.5
30.1
80.3
50.2
8YES
TL
NO
29
3[1
12
]0.0
50.1
50.0
90.0
60.1
50.0
90.0
60.1
50.0
90.0
6NO
0.0
5YES
30
3[1
11
]0.0
30.0
40.0
40.0
40.0
40.0
40.0
40.0
40.0
40.0
4NO
0.0
3YES
31
3[2
21
]0.5
37.0
40.5
80.5
50.5
80.3
70.5
70.5
80.4
80.6
0YES
TL
YES
32
3[1
13
]0.1
10.1
90.1
40.1
50.1
00.1
40.1
40.1
90.1
50.1
5YES
0.1
1YES
33
3[3
23
]0.0
70.1
50.0
80.0
80.1
50.0
80.0
80.1
50.0
80.0
8NO
0.0
7TL
34
3[1
22
]0.4
20.1
50.1
50.1
50.1
50.1
50.1
50.1
50.1
50.1
5YES
1.0
3TL
35
3[2
22
]0.0
60.1
40.0
70.0
70.1
40.0
70.0
70.1
40.0
70.0
7NO
0.0
6NO
36
3[1
13
]0.5
72.6
92.6
46.3
70.6
719.0
23.3
31.6
8316.9
40.6
8YES
TL
TL
37
3[2
13
]0.2
50.2
00.2
50.2
50.3
40.2
00.2
00.3
50.1
90.2
6YES
0.1
5NO
38
3[2
22
]0.0
60.2
00.1
00.0
70.2
00.1
00.0
70.2
00.1
00.0
7NO
0.0
6NO
39
3[3
31
]0.0
70.1
60.0
80.0
80.1
60.0
80.0
80.1
60.0
80.0
8NO
0.0
7TL
40
3[2
12
]0.1
30.2
40.1
80.1
80.1
50.1
70.1
70.1
90.1
70.1
8YES
0.1
4YES
41
3[2
23
]0.0
80.2
10.1
40.0
80.2
20.1
40.0
90.2
20.1
40.0
9NO
0.0
7NO
42
3[1
11
]0.0
30.0
40.0
40.0
40.0
40.0
40.0
40.0
40.0
40.0
4NO
0.0
3NO
43
3[1
32
]1.0
20.1
00.1
00.1
00.1
00.1
00.1
00.1
00.1
00.1
0YES
TL
YES
44
3[3
22
]0.1
00.4
20.1
80.2
00.4
20.1
80.2
00.4
20.1
80.1
9NO
0.1
0NO
45
3[2
21
]0.0
50.1
30.0
60.0
60.1
30.0
60.0
60.1
30.0
60.0
6NO
0.0
5TL
46
3[1
22
]0.0
50.1
20.0
60.0
60.1
20.0
60.0
60.1
20.0
60.0
6NO
0.0
5NO
47
3[1
32
]0.4
00.0
30.0
30.0
30.0
30.0
30.0
30.0
30.0
30.0
3YES
0.5
8NO
48
3[2
21
]0.0
40.0
50.0
50.0
50.0
50.0
50.0
50.0
50.0
50.0
5NO
0.0
4YES
49
3[2
11
]0.0
40.1
10.0
70.0
50.1
10.0
70.0
50.1
10.0
70.0
5NO
0.0
4NO
50
4[1
21
2]
0.1
80.0
60.0
60.0
60.0
60.0
60.0
60.0
60.0
60.0
6YES
0.1
7NO
51
4[1
11
1]
0.0
60.1
10.0
70.0
70.1
10.0
70.0
70.1
10.0
70.0
7NO
0.0
6NO
52
4[3
13
1]
0.1
00.1
70.1
10.1
10.1
70.1
10.1
10.1
80.1
10.1
1NO
0.1
0NO
53
4[3
12
1]
775.5
50.3
60.3
60.3
60.3
60.3
60.3
50.3
60.3
60.3
7YES
TL
TL
54
4[1
12
3]
0.0
90.2
20.1
00.1
00.2
20.1
00.1
00.2
20.1
00.1
0NO
0.0
9NO
55
4[1
21
2]
0.1
40.4
30.3
60.3
60.1
60.1
80.1
80.1
60.1
80.3
6YES
0.1
7NO
56
4[2
21
2]
1.6
90.2
00.2
00.2
00.2
00.2
10.2
00.2
00.2
00.2
0YES
TL
NO
57
4[1
22
2]
0.2
90.4
20.2
80.2
80.2
10.3
10.3
10.4
80.4
50.3
1YES
0.6
4YES
58
4[2
22
1]
0.0
90.2
40.1
10.1
10.2
40.1
10.1
10.2
40.1
10.1
1NO
0.0
9YES
59
4[1
22
1]
0.0
90.2
20.1
10.1
10.2
30.1
10.1
10.2
20.1
10.1
1NO
0.0
9NO
60
4[1
31
3]
0.0
90.1
60.1
10.1
10.1
60.1
10.1
10.1
60.1
10.1
1NO
0.1
0YES
61
4[3
13
2]
38.8
3TL
TL
17.7
00.3
12.6
454.9
994.3
81.3
6TL
YES
152.4
8NO
62
4[1
13
2]
24.3
00.8
00.7
90.7
90.7
90.8
00.8
00.7
90.8
00.8
0YES
TL
NO
63
4[2
32
3]
0.3
20.7
00.4
10.4
10.2
40.3
60.3
60.4
70.4
10.4
2YES
0.2
5NO
64
4[2
23
1]
0.1
80.7
60.3
20.3
40.7
70.3
20.3
40.7
60.3
20.3
4NO
0.1
8NO
65
4[2
13
2]
1.3
22.6
32.2
51.8
90.4
21.2
61.2
73.0
61.8
01.6
4YES
1.6
1NO
66
4[3
33
3]
0.5
81.3
40.7
70.8
20.3
70.8
80.6
00.6
60.5
70.5
0YES
0.4
4YES
32
67
4[3
22
1]
0.1
20.4
40.2
20.1
40.4
40.2
20.1
30.4
40.2
20.1
4NO
0.1
2NO
68
4[3
22
2]
1.9
00.2
00.2
00.2
00.2
00.2
00.2
00.2
00.2
00.2
0YES
TL
YES
69
4[2
13
3]
0.1
20.2
80.1
30.1
30.2
80.1
30.1
30.2
80.1
30.1
3NO
0.1
2NO
70
4[2
22
1]
0.0
80.1
60.1
00.1
00.1
60.1
00.1
00.1
60.1
00.1
0NO
0.0
9YES
71
4[1
22
2]
0.0
80.0
90.0
90.0
90.0
90.0
90.0
90.0
90.0
90.0
9NO
0.0
8YES
72
4[1
21
3]
0.1
00.3
20.2
00.1
20.3
30.2
00.1
20.3
30.2
00.1
2NO
0.1
0NO
73
4[2
22
2]
160.8
212.6
612.8
012.7
112.6
512.8
112.6
412.7
512.8
312.6
6YES
TL
NO
74
4[2
11
3]
0.1
60.8
50.3
70.2
80.8
50.3
70.2
80.8
50.3
90.2
8NO
0.1
6NO
75
4[1
21
3]
0.1
00.2
40.1
10.1
10.2
40.1
10.1
10.2
40.1
10.1
1NO
0.1
0YES
76
4[1
11
2]
0.0
60.1
10.0
70.0
70.1
10.0
70.0
70.1
10.0
70.0
7NO
0.0
6YES
77
4[3
12
2]
0.1
40.5
20.2
60.1
60.5
20.2
60.1
60.5
20.2
60.1
6NO
0.1
4NO
78
4[2
11
2]
0.2
40.3
40.2
80.2
80.2
10.2
80.2
80.2
10.2
80.2
8YES
0.2
5YES
79
4[3
21
3]
54.9
9297.6
5297.3
4296.3
74.6
8415.6
4422.4
235.0
457.3
261.4
7YES
TL
NO
80
4[2
11
2]
0.0
70.0
80.0
80.0
80.0
80.0
80.0
80.0
80.0
80.0
8NO
0.0
7YES
81
4[3
22
2]
0.1
90.8
00.3
30.3
60.8
10.3
30.3
70.8
10.3
30.3
7NO
0.1
9YES
82
4[2
31
1]
0.2
30.6
40.3
20.2
60.1
70.2
20.2
60.2
30.1
90.2
6YES
0.4
0NO
83
4[2
12
2]
TL
1.5
01.3
6TL
8.7
351.2
8TL
8.7
476.2
61209.7
6YES
0.8
8NO
84
4[3
31
2]
0.2
61.8
90.7
90.6
61.9
00.8
00.6
61.9
10.7
90.6
6NO
0.2
6NO
85
4[1
11
2]
0.0
70.1
20.0
80.0
80.1
20.0
80.0
80.1
20.0
80.0
8NO
0.0
7YES
86
4[1
11
3]
0.1
50.1
50.1
50.1
50.1
50.1
50.1
50.1
50.1
40.1
5YES
0.2
1NO
87
4[2
31
2]
0.4
234.4
434.3
10.4
60.7
40.4
20.4
40.5
8TL
0.4
9YES
TL
NO
88
4[1
22
1]
0.1
80.3
30.2
20.2
20.1
80.2
10.2
20.2
50.2
10.2
2YES
0.1
8YES
89
4[1
12
1]
0.0
50.0
60.0
60.0
60.0
60.0
60.0
60.0
60.0
60.0
6NO
0.0
5NO
90
4[3
32
1]
0.4
70.5
10.3
30.3
30.2
40.3
20.3
10.5
20.3
40.3
2YES
0.7
1NO
91
4[3
13
2]
TL
0.3
80.3
80.3
80.3
80.3
80.3
80.3
80.3
80.3
8YES
TL
NO
92
4[3
32
3]
0.4
60.9
20.7
30.8
60.2
30.5
70.4
80.4
60.3
60.4
0YES
1.0
0YES
93
4[2
22
2]
0.4
10.6
60.5
10.4
00.6
00.4
40.4
60.7
00.8
50.4
3YES
0.3
2NO
94
4[2
23
2]
1.0
03.2
82.0
32.8
60.4
41.7
41.2
33.3
157.7
22.6
5YES
15.2
4NO
95
4[1
12
2]
0.0
80.2
00.0
90.0
90.2
00.0
90.0
90.2
00.0
90.0
9NO
0.0
8YES
96
4[1
21
1]
0.0
60.1
10.0
70.0
70.1
10.0
70.0
70.1
10.0
70.0
7NO
0.0
6YES
97
4[2
22
2]
0.2
20.3
20.2
60.2
60.2
20.2
50.2
50.3
20.2
50.2
6YES
0.2
2NO
98
4[1
12
1]
0.2
70.1
80.1
80.1
80.1
80.1
80.1
80.1
80.1
80.1
8YES
98.1
4NO
99
4[2
22
2]
0.1
70.6
30.3
20.1
90.6
40.3
20.1
90.6
40.3
20.1
9NO
0.1
7YES
100
5[2
22
11
]0.1
50.3
80.1
70.1
60.3
90.1
60.1
70.3
80.1
70.1
7NO
0.1
5NO
101
5[2
33
22
]TL
TL
TL
TL
1.8
2TL
TL
TL
TL
TL
YES
TL
YES
102
5[2
22
33
]4.4
52.3
61.1
41.3
64.7
12.8
441.9
751.4
71.3
53.7
6YES
TL
YES
103
5[1
23
12
]1.2
90.2
80.2
80.2
80.2
80.2
80.2
80.2
80.2
80.2
8YES
TL
YES
104
5[1
31
11
]0.3
70.3
10.3
30.3
30.4
20.3
10.3
10.6
00.3
10.3
3YES
0.2
3NO
105
5[2
32
21
]0.3
21.7
60.7
70.5
81.7
80.7
70.5
81.7
80.7
70.5
8NO
0.3
2NO
106
5[2
21
22
]0.1
90.4
80.2
10.2
10.4
80.2
00.2
00.4
80.2
00.2
0NO
0.1
9NO
107
5[1
23
21
]0.2
10.8
10.3
80.2
30.8
30.3
80.2
30.8
30.3
80.2
3NO
0.2
1NO
108
5[3
22
11
]0.1
40.3
50.1
70.1
60.3
50.1
70.1
60.3
60.1
60.1
7NO
0.1
5YES
109
5[2
22
31
]582.9
42.1
22.0
92.1
12.1
12.1
22.1
12.1
02.1
12.1
0YES
TL
YES
110
5[3
22
33
]0.2
90.6
70.3
30.3
20.6
90.3
20.3
20.6
70.3
20.3
2NO
0.3
0YES
111
5[2
13
13
]0.2
10.6
60.4
10.2
30.6
60.4
10.2
30.6
60.4
10.2
3NO
0.2
1NO
112
5[1
21
32
]0.4
90.7
40.4
60.5
00.3
20.4
30.5
10.7
50.4
30.5
2YES
1.1
1YES
113
5[1
12
11
]0.3
40.7
60.3
80.3
80.2
50.3
80.3
80.5
10.3
80.3
8YES
0.5
3YES
114
5[1
33
11
]0.2
10.5
30.2
30.2
30.5
30.2
30.2
30.5
30.2
30.2
3NO
0.2
2NO
115
5[3
12
32
]0.2
40.5
90.2
60.2
60.5
90.2
60.2
60.5
90.2
60.2
7NO
0.2
4YES
116
5[1
22
32
]0.1
60.3
00.1
80.1
80.3
00.1
80.1
80.3
00.1
80.1
8NO
0.1
6NO
117
5[2
21
22
]6.0
720.3
65.1
85.0
11.0
91.7
92.8
10.3
427.8
13.0
9YES
4.2
0NO
118
5[2
22
22
]0.2
00.4
70.2
20.2
20.4
80.2
20.2
20.4
80.2
20.2
2NO
0.2
0NO
119
5[3
32
11
]1.6
12.0
325.4
96.8
80.6
81.8
61.3
412.0
03.3
547.5
3YES
TL
NO
120
5[3
21
22
]0.1
90.5
00.2
20.2
10.5
00.2
20.2
20.4
90.2
20.2
2NO
0.2
0NO
121
5[1
21
21
]0.1
40.3
50.1
60.1
50.3
60.1
50.1
60.3
60.1
60.1
5NO
0.1
3YES
122
5[1
22
23
]0.2
00.5
10.2
30.2
30.5
10.2
30.2
30.5
20.2
30.2
3NO
0.2
1NO
123
5[2
22
32
]0.2
10.5
30.2
30.2
30.5
30.2
30.2
30.5
30.2
30.2
4NO
0.2
1NO
124
5[1
12
13
]2.1
00.2
60.2
60.2
60.2
60.2
60.2
60.2
60.2
60.2
6YES
TL
NO
125
5[2
32
12
]2.1
20.5
00.5
10.5
00.4
90.5
00.5
00.5
00.5
10.5
1YES
TL
NO
126
5[2
22
22
]0.9
50.4
70.4
70.4
70.4
70.4
70.4
70.4
70.4
80.4
8YES
TL
YES
127
5[2
12
12
]0.3
00.5
20.3
30.3
30.2
20.3
00.3
00.5
10.2
90.3
3YES
0.2
3YES
128
5[2
12
23
]TL
0.4
80.4
80.4
80.4
80.4
80.4
80.4
80.4
80.4
8YES
TL
YES
129
5[2
23
33
]TL
TL
TL
TL
204.9
5TL
TL
TL
TL
83.4
6YES
TL
NO
130
5[3
21
22
]TL
TL
TL
TL
65.0
1TL
TL
74.4
2TL
TL
YES
TL
NO
131
5[2
12
21
]0.2
10.6
80.4
00.2
30.6
80.4
10.2
30.6
80.4
10.2
3NO
0.2
1YES
132
5[2
21
12
]0.1
90.4
80.2
10.2
10.4
80.2
10.2
10.4
80.2
10.2
1NO
0.1
9TL
133
5[2
22
23
]TL
0.4
50.4
50.4
50.4
50.4
50.4
50.4
50.4
50.4
5YES
TL
NO
134
5[2
21
22
]5.1
61.0
40.7
51.1
20.7
2170.7
3TL
0.7
2TL
TL
YES
0.8
7NO
135
5[2
21
21
]0.1
40.3
60.1
60.1
60.3
60.1
60.1
60.3
60.1
60.1
6NO
0.1
4YES
136
5[2
21
32
]0.2
40.9
40.4
50.2
60.9
40.4
50.2
60.9
40.4
50.2
6NO
0.2
4YES
137
5[2
22
21
]0.4
00.8
80.6
10.6
10.3
50.4
80.4
70.6
50.4
50.4
5YES
0.2
9YES
138
5[2
22
21
]0.2
61.3
70.6
90.4
91.3
80.6
90.4
91.3
70.6
90.4
9NO
0.2
7NO
139
5[3
21
23
]TL
0.5
80.5
80.5
80.5
80.5
80.5
80.5
80.5
80.5
8YES
TL
NO
140
5[2
21
12
]0.1
10.2
10.1
30.1
30.2
10.1
30.1
30.2
20.1
30.1
3NO
0.1
1YES
141
5[1
21
21
]0.6
20.9
41.1
30.6
70.3
30.5
80.7
10.7
20.9
50.9
9YES
1.1
5TL
142
5[2
12
21
]0.2
10.8
30.3
90.2
30.8
40.3
90.2
20.8
30.3
90.2
2NO
0.2
1YES
143
5[1
21
22
]0.1
90.4
60.2
00.2
10.4
60.2
00.2
00.4
60.2
00.2
0NO
0.1
9YES
33
144
5[1
31
22
]0.1
70.3
10.1
90.1
90.3
10.1
90.1
90.3
10.1
90.1
9NO
0.1
7YES
145
5[1
22
21
]0.1
60.4
20.1
80.1
80.4
20.1
80.1
80.4
20.1
80.1
8NO
0.1
7NO
146
5[3
22
12
]TL
3.1
7688.6
720.6
0203.0
7TL
TL
2.2
7TL
21.6
6YES
TL
NO
147
5[1
12
23
]0.1
60.2
90.1
80.1
80.3
00.1
80.1
80.3
00.1
80.1
8NO
0.1
6YES
148
5[2
22
23
]0.2
50.6
20.2
80.2
70.6
20.2
70.2
80.6
20.2
70.2
7NO
0.2
5YES
149
5[2
11
22
]0.1
30.3
50.1
50.1
50.3
50.1
50.1
50.3
50.1
50.1
5NO
0.1
3NO
Tab
le5:
MN
Ean
dP
NE
resu
lts
for
Inst
an
ceS
etB
.S
am
en
ota
tion
as
Tab
le4.
#L
FFE
seq
seq3
seq5
rseq1
rseq3
rseq5
rand1
rand3
rand5
MNE
FE-P
PNE
07
[1
12
22
22
]TL
TL
TL
TL
TL
8.8
6TL
240.9
9TL
TL
YES
TL
TL
17
[1
11
21
21
]62.1
22.8
21.9
3102.9
71.6
46.9
1279.0
61.8
75.9
883.8
4YES
TL
TL
27
[2
32
13
22
]TL
6.2
86.2
96.3
26.4
36.4
26.3
56.2
86.3
56.3
2YES
TL
TL
37
[1
22
11
31
]572.6
90.5
10.5
00.5
00.5
10.5
00.5
00.5
00.5
00.5
0YES
TL
TL
47
[1
32
21
21
]TL
1.3
41.3
31.3
31.3
31.3
51.3
21.3
31.3
51.3
2YES
TL
TL
57
[2
12
21
22
]192.3
70.2
80.2
70.2
80.2
80.2
80.2
70.2
70.2
70.2
7YES
2.4
0YES
67
[2
22
33
21
]TL
TL
TL
TL
20.0
2TL
TL
TL
TL
TL
YES
TL
TL
77
[2
21
13
23
]TL
TL
TL
TL
464.2
1TL
TL
TL
TL
TL
YES
TL
TL
87
[3
31
13
33
]TL
0.3
80.3
80.3
90.3
80.3
80.3
90.3
80.3
90.3
9YES
TL
TL
97
[2
12
22
12
]TL
2.4
52.4
62.4
52.4
32.4
72.4
62.4
72.4
62.4
6YES
TL
TL
10
7[2
32
21
22
]TL
TL
TL
TL
1704.4
2TL
TL
TL
TL
TL
YES
TL
TL
11
7[1
23
13
21
]TL
TL
418.9
6TL
2.8
18.9
1TL
4.1
7TL
TL
YES
TL
TL
12
7[3
22
12
11
]9.2
93.6
72.7
08.1
71.3
84.5
9TL
4.8
69.1
43.9
0YES
TL
TL
13
7[2
33
11
12
]TL
251.7
8250.4
864.7
9687.6
3TL
22.4
5TL
21.4
331.4
6YES
15.6
6YES
14
7[3
22
22
22
]TL
TL
TL
TL
234.3
3TL
TL
480.5
6TL
TL
YES
TL
TL
15
7[2
23
22
21
]10.6
52.0
91.1
216.3
30.6
919.9
14.1
35.1
15.2
518.6
8YES
2.2
4YES
16
7[2
22
21
22
]TL
162.8
9162.1
9TL
14.1
451.7
5TL
82.2
1975.4
9TL
YES
TL
TL
17
7[3
31
32
21
]635.2
7TL
TL
TL
1.8
1TL
TL
6.8
892.2
9TL
YES
TL
TL
18
7[1
12
22
22
]TL
TL
TL
TL
TL
8.7
9TL
TL
10.9
2269.5
7YES
TL
TL
19
7[2
11
32
32
]0.9
78.9
93.5
72.2
89.0
03.5
72.3
08.9
63.5
92.3
0NO
0.9
7NO
20
7[2
23
22
21
]10.7
72.1
21.1
216.3
20.7
020.1
54.0
82.5
64.7
25.8
2YES
2.2
4YES
21
7[3
22
22
22
]TL
TL
TL
TL
231.0
1TL
TL
TL
TL
TL
YES
TL
TL
22
7[2
11
23
13
]1.2
710.2
94.1
93.3
910.3
34.1
63.4
010.3
54.1
63.4
1NO
1.2
6NO
23
7[2
33
11
12
]TL
247.8
0248.0
164.2
8674.3
7TL
22.2
6TL
56.8
9TL
YES
15.7
3YES
24
7[1
12
22
22
]TL
TL
TL
TL
TL
8.8
5TL
40.1
4TL
72.5
2YES
TL
TL
25
7[3
31
32
21
]634.3
0TL
TL
TL
1.8
2TL
TL
21.9
247.9
7TL
YES
TL
TL
26
7[2
22
21
22
]TL
163.1
3162.7
7TL
14.1
351.8
4TL
4.8
249.2
3TL
YES
TL
TL
27
7[2
23
22
21
]10.8
02.1
01.1
216.5
00.7
020.2
04.1
11.3
23.9
629.0
7YES
2.2
5YES
28
7[3
22
22
22
]TL
TL
TL
TL
232.5
6TL
TL
TL
TL
TL
YES
TL
TL
29
7[2
33
11
12
]TL
251.4
1250.9
264.2
4687.3
8TL
22.5
2TL
26.5
52.1
1YES
15.8
3YES
30
7[3
22
12
11
]9.3
23.6
32.7
18.1
71.3
74.6
2TL
3.8
6TL
2.7
7YES
TL
TL
31
7[1
23
13
21
]TL
TL
419.4
8TL
2.8
08.8
8TL
3.5
51403.1
9TL
YES
TL
TL
32
7[2
32
21
22
]TL
TL
TL
TL
1687.5
3TL
TL
TL
TL
TL
YES
TL
TL
33
7[2
12
22
12
]TL
2.4
62.4
62.4
52.4
62.4
62.4
32.4
42.4
72.4
3YES
TL
TL
34
7[3
31
13
33
]TL
0.3
80.3
90.3
80.3
90.3
80.3
80.3
80.3
80.3
8YES
TL
TL
35
7[2
21
13
23
]TL
TL
TL
TL
458.7
2TL
TL
TL
TL
TL
YES
TL
TL
36
7[2
22
33
21
]TL
TL
TL
TL
20.1
5TL
TL
TL
TL
TL
YES
TL
TL
37
7[2
12
21
22
]194.2
80.2
70.2
80.2
80.2
90.2
80.2
70.2
80.2
80.2
8YES
2.4
9YES
38
7[1
32
21
21
]TL
1.3
21.3
21.3
31.3
21.3
31.3
31.3
21.3
31.3
3YES
TL
TL
39
7[1
22
11
31
]572.4
20.5
00.5
00.5
00.5
00.5
00.5
00.5
00.5
00.5
0YES
TL
TL
40
7[2
32
13
22
]TL
6.3
16.3
66.3
16.3
46.3
26.3
06.3
66.3
26.3
1YES
TL
TL
41
7[1
11
21
21
]62.5
52.7
91.9
2103.2
81.6
16.8
8281.1
88.2
923.7
8TL
YES
TL
TL
42
7[1
23
32
13
]TL
4.9
64.9
94.9
44.9
85.0
04.9
54.9
54.9
44.9
7YES
TL
TL
43
7[2
22
22
21
]TL
9.4
79.5
89.5
09.4
89.5
99.5
19.5
39.5
59.5
6YES
TL
TL
44
7[3
11
22
22
]328.7
223.7
322.8
0151.0
336.6
336.9
11190.5
511.5
9168.2
04.9
8YES
TL
TL
45
7[2
22
31
22
]62.3
90.1
20.1
30.1
20.1
20.1
20.1
20.1
30.1
30.1
2YES
TL
TL
46
7[2
12
22
32
]TL
TL
TL
TL
357.4
3TL
TL
TL
TL
TL
YES
TL
TL
47
7[2
22
11
32
]1131.2
51.6
71.6
91.6
61.6
71.6
61.6
61.6
91.6
61.6
6YES
TL
TL
48
7[1
22
22
32
]72.6
819.0
918.0
618.0
8130.4
7136.2
1136.3
747.9
7173.2
2113.8
7YES
TL
TL
49
7[2
22
12
33
]113.3
0TL
TL
TL
4.8
5TL
115.9
6413.0
11065.6
7689.1
0YES
TL
TL
34
Tab
le6:
Inst
ance
s’so
luti
ons
for
Inst
an
ceS
etIn
sigh
ts.
The
colu
mns
are,
inor
der
ofap
pea
rance
:th
ein
stan
ce’s
num
ber
,th
eb
ool
ean
tax
swit
ch(T
a)
an
dth
etr
ad
esw
itch
(Tr).
Th
en,
the
set
of
resu
lts
ass
oci
ate
dw
ith
each
of
the
two
cou
ntr
ies
(Cou
ntr
yO
ne,
an
dC
ou
ntr
yT
wo
).In
part
icu
lar:
the
un
it-e
ner
gypro
du
ctio
nle
vel
Pro
d,
the
dom
esti
cp
rice
per
un
it-e
ner
gy$(
E),
the
imp
ort
Imp
and
exp
ort
Exp
un
it-e
ner
gies
,th
eex
por
tp
rice
$(E
),an
dth
eta
xp
erunit
-em
issi
onT
ax.
Furt
her
mor
e,fo
rea
chof
the
the
3fo
llow
ers
ofea
chco
untr
y,w
ehav
eth
ety
pe
Ty
(Cfo
rco
al,G
for
gas,
orS
for
sola
r),
the
asso
ciat
edem
issi
onco
stp
eru
nit
-en
ergy
E,
an
dit
sp
rod
uct
ion
Pro
d.
Country
One
Country
Tw
oFollow
er
1Follow
er
2Follow
er
3Follow
er
1Follow
er
2Follow
er
3#
Ta
Tr
Prod
$(D
)Im
pExp
$(E)
Tax
Ty
EP
rod
TE
Prod
Ty
EP
rod
Prod
$(D
)Im
pExp
$(E)
Tax
Ty
EP
rod
TE
Prod
Ty
EP
rod
I0
00
83,3
3300,0
00,0
00,0
0-
80,0
0C
500
0,0
0G
200
41,6
7S
25
41,6
760,0
0255,0
00,0
00,0
0-
0,3
0C
500
27,7
6G
100
27,7
6S
50
4,4
8I0
01
98,8
1300,0
00,5
616,0
3279,4
380,0
0C
500
0,0
0G
200
57,1
4S
25
41,6
744,5
2255,0
016,0
30,5
6278,4
39,3
5C
500
20,5
2G
100
20,5
2S
50
3,4
9I0
10
83,3
3300,0
00,0
00,0
0-
0,1
4C
500
7,3
5G
200
37,2
2S
25
38,7
660,0
0255,0
00,0
00,0
0-
0,0
0C
500
27,4
2G
100
27,8
8S
50
4,6
9I0
11
119,3
2300,0
00,0
035,9
9103255,8
60,0
7C
500
32,0
2G
200
47,1
0S
25
40,2
024,0
1255,0
035,9
90,0
0103256,8
60,0
7C
500
0,1
4G
100
22,4
3S
50
1,4
5I1
00
37,5
0270,0
00,0
00,0
0-
11,4
1C
300
29,6
9G
100
7,8
1S
50
0,0
050,0
0315,0
00,0
00,0
0-
15,7
7G
100
26,9
2S
50
11,5
4S
25
11,5
4I1
01
0,0
0270,0
046,5
99,0
963,5
450,0
0C
300
0,0
0G
100
0,0
0S
50
0,0
087,5
0315,0
09,0
946,5
962,5
47,2
7G
100
36,3
6S
50
25,5
7S
25
25,5
7I1
10
37,5
0270,0
00,0
00,0
0-
0,0
6C
300
24,7
3G
100
12,7
7S
50
0,0
050,0
0315,0
00,0
00,0
0-
0,2
8G
100
13,5
6S
50
13,8
8S
25
22,5
6I1
11
3,8
3270,0
033,6
70,0
05498,0
30,1
6C
300
0,5
7G
100
3,2
6S
50
0,0
083,6
7315,0
00,0
033,6
75497,0
30,1
1G
100
31,8
2S
50
24,1
5S
25
27,7
0I2
00
30,5
6247,5
00,0
00,0
0-
15,5
6C
300
22,0
3G
100
8,5
3S
25
0,0
097,5
0276,2
50,0
00,0
0-
20,0
4C
500
53,5
3G
200
36,2
1S
50
7,7
6I2
01
52,4
0247,5
030,0
351,8
8304,1
00,0
0C
300
32,7
6G
100
19,6
4S
25
0,0
075,6
5276,2
551,8
830,0
3303,1
040,5
7C
500
33,9
8G
200
26,3
2S
50
15,3
5I2
10
30,5
6247,5
00,0
00,0
0-
0,0
8C
300
16,5
2G
100
14,0
4S
25
0,0
097,5
0276,2
50,0
00,0
0-
0,0
9C
500
31,2
4G
200
38,8
7S
50
27,3
8I2
11
52,4
0247,5
00,0
021,8
5161540,0
60,0
0C
300
32,7
6G
100
19,6
4S
25
0,0
075,6
5276,2
521,8
50,0
0161541,0
60,1
2C
500
17,1
6G
200
32,9
6S
50
25,5
3I3
00
84,3
7382,5
00,0
00,0
0-
116,8
8C
500
35,0
9G
200
35,0
9S
25
14,2
030,5
6247,5
00,0
00,0
0-
32,7
3S
25
10,1
9S
50
10,1
9S
50
10,1
9I3
01
16,6
5382,5
079,6
211,9
0178,8
5153,3
4C
500
2,8
2G
200
7,0
5S
25
6,7
998,2
8247,5
011,9
079,6
2179,8
50,0
0S
25
32,7
6S
50
32,7
6S
50
32,7
6I3
10
134,7
1342,2
30,0
00,0
0-
0,2
4C
500
0,0
0G
200
56,4
2S
25
78,2
954,2
2226,2
00,0
00,0
0-
0,0
0S
25
18,0
7S
50
18,0
7S
50
18,0
7I3
11
178,2
8374,0
414,9
698,3
064,8
70,3
1C
500
0,0
0G
200
71,1
0S
25
107,1
80,0
0200,0
098,3
014,9
665,8
70,0
0S
25
0,0
0S
50
0,0
0S
50
0,0
0I4
00
108,4
3277,4
10,0
00,0
0-
25,0
0S
25
36,1
4S
50
36,1
4S
50
36,1
493,7
5318,7
50,0
00,0
0-
0,2
9G
200
11,6
8S
50
36,4
9S
25
45,5
9I4
01
151,6
4298,2
90,0
066,4
127,1
125,0
0S
25
50,5
5S
50
50,5
5S
50
50,5
527,3
4318,7
566,4
10,0
028,1
10,8
8G
200
0,0
0S
50
0,0
0S
25
27,3
4I4
10
93,5
5290,8
00,0
00,0
0-
1,0
9S
25
43,7
6S
50
24,9
0S
50
24,9
093,7
5318,7
50,0
00,0
0-
0,2
9G
200
11,6
8S
50
36,4
9S
25
45,5
9I4
11
101,4
2298,4
40,0
016,3
6159,0
91,1
9S
25
47,4
4S
50
26,9
9S
50
26,9
977,3
9318,7
516,3
60,0
0160,0
90,3
4G
200
0,1
7S
50
33,2
5S
25
43,9
7I5
00
112,5
0318,7
50,0
00,0
0-
110,9
6C
300
52,5
4G
200
52,5
4S
25
7,4
280,0
0240,0
00,0
00,0
0-
300,0
0C
500
0,0
0G
200
49,2
3G
200
30,7
7I5
01
192,5
0318,7
560,9
3140,9
3188,8
691,3
9C
300
70,3
3G
200
70,3
3S
25
51,8
50,0
0240,0
0140,9
360,9
3189,8
690,0
0C
500
0,0
0G
200
0,0
0G
200
0,0
0I5
10
136,6
4306,6
80,0
00,0
0-
0,5
2C
300
0,0
0G
200
47,4
8S
25
89,1
680,0
0240,0
00,0
00,0
0-
0,1
3C
500
20,3
4G
200
48,1
4G
200
11,5
3I5
11
150,8
4318,7
50,0
038,3
4116,2
20,5
6C
300
0,0
0G
200
51,1
4S
25
99,7
041,6
6240,0
038,3
40,0
0117,2
20,1
8C
500
0,0
0G
200
39,3
1G
200
2,3
6I6
00
56,2
5255,0
00,0
00,0
0-
0,0
0C
500
25,3
0G
100
26,6
0S
50
4,3
581,2
5276,2
50,0
00,0
0-
9,0
1G
200
42,9
4S
50
19,1
5S
50
19,1
5I6
01
28,0
3255,0
028,2
20,0
0398,9
70,0
6C
500
3,8
9G
100
22,3
2S
50
1,8
2109,4
7276,2
50,0
028,2
2397,9
70,0
0G
200
51,1
4S
50
29,1
7S
50
29,1
7I6
10
56,2
5255,0
00,0
00,0
0-
0,0
0C
500
25,3
0G
100
26,6
0S
50
4,3
581,2
5276,2
50,0
00,0
0-
0,1
0G
200
33,6
2S
50
23,8
1S
50
23,8
1I6
11
28,0
3255,0
028,2
20,0
0190054,1
40,0
6C
500
3,8
9G
100
22,3
2S
50
1,8
2109,4
7276,2
50,0
028,2
2190053,1
40,0
0G
200
51,1
4S
50
29,1
7S
50
29,1
7I7
00
58,3
3297,5
00,0
00,0
0-
103,7
5C
300
29,1
7G
100
29,1
7S
25
0,0
056,5
2399,1
30,0
00,0
0-
199,1
3C
500
0,0
0G
100
0,0
0G
200
56,5
2I7
01
0,0
0297,5
058,3
30,0
099,0
0147,5
0C
300
0,0
0G
100
0,0
0S
25
0,0
0108,3
3405,0
00,0
058,3
3100,0
0205,0
0C
500
0,0
0G
100
47,6
2G
200
60,7
1I7
10
91,9
4267,2
50,0
00,0
0-
0,3
9C
300
0,1
3G
100
52,1
6S
25
39,6
5122,6
1339,6
50,0
00,0
0-
0,2
8C
500
0,0
0G
100
77,0
5G
200
45,5
6I7
11
124,5
6297,5
00,0
066,2
2114,1
80,4
9C
300
0,1
7G
100
65,6
1S
25
58,7
891,5
7307,9
866,2
20,0
0115,1
80,2
2C
500
0,0
0G
100
59,5
8G
200
31,9
9I8
00
150,0
0360,0
00,0
00,0
0-
2,4
6S
50
32,1
7S
25
85,6
5S
50
32,1
760,9
4276,2
50,0
00,0
0-
58,9
1C
500
48,1
0G
200
12,8
4S
50
0,0
0I8
01
210,9
4360,0
00,0
060,9
445,0
01,9
0S
50
56,5
5S
25
97,8
4S
50
56,5
50,0
0276,2
560,9
40,0
046,0
0126,2
5C
500
0,0
0G
200
0,0
0S
50
0,0
0I8
10
150,0
0360,0
00,0
00,0
0-
2,4
6S
50
32,1
7S
25
85,6
5S
50
32,1
760,9
4276,2
50,0
00,0
0-
0,2
4C
500
5,4
6G
200
21,3
4S
50
34,1
5I8
11
157,5
0360,0
00,0
07,5
01045,0
02,3
9S
50
35,1
8S
25
87,1
5S
50
35,1
853,4
3276,2
57,5
00,0
01046,0
00,2
5C
500
0,5
3G
200
19,2
9S
50
33,6
1I9
00
150,0
0360,0
00,0
00,0
0-
2,3
0C
300
0,0
0S
25
75,0
0S
25
75,0
093,3
3280,0
00,0
00,0
0-
0,4
7G
100
25,4
7S
25
38,6
2S
50
29,2
4I9
01
219,3
3360,0
00,0
069,3
325,0
00,7
7C
300
0,0
0S
25
109,6
7S
25
109,6
724,0
0280,0
069,3
30,0
026,0
01,2
0G
100
0,0
0S
25
24,0
0S
50
0,0
0I9
10
190,1
5335,9
10,0
00,0
0-
0,4
5C
300
0,0
0S
25
95,0
8S
25
95,0
8100,4
5274,6
60,0
00,0
0-
0,3
2G
100
32,4
6S
25
37,2
4S
50
30,7
5I9
11
230,4
7360,0
09,1
589,6
23239,8
10,5
3C
300
0,1
4S
25
115,1
6S
25
115,1
660,6
6244,1
589,6
29,1
53238,8
10,1
9G
100
19,2
0S
25
22,6
5S
50
18,8
1I10
00
69,5
7283,2
60,0
00,0
0-
83,2
6G
100
0,0
0S
25
34,7
8S
50
34,7
837,5
0270,0
00,0
00,0
0-
11,4
1C
300
29,6
9G
100
7,8
1S
50
0,0
0I10
01
91,6
7292,5
00,0
037,5
025,0
092,5
0G
100
0,0
0S
25
48,4
8S
50
43,1
8-0
,00
270,0
037,5
00,0
026,0
050,0
0C
300
-0,0
0G
100
0,0
0S
50
0,0
0I10
10
67,0
9284,7
50,0
00,0
0-
0,8
0G
100
4,0
1S
25
40,6
5S
50
22,4
337,5
0270,0
00,0
00,0
0-
0,0
6C
300
24,7
3G
100
12,7
7S
50
0,0
0I10
11
87,8
4292,5
00,0
033,6
711284,5
50,8
0G
100
10,7
0S
25
47,6
8S
50
29,4
63,8
3270,0
033,6
70,0
011285,5
50,1
6C
300
0,5
7G
100
3,2
6S
50
0,0
0I11
00
125,7
6286,9
70,0
00,0
0-
25,0
0S
50
34,9
7S
50
34,9
7S
25
55,8
175,0
0337,5
00,0
00,0
0-
0,5
8C
500
0,0
0G
200
1,7
9S
50
73,2
1I11
01
173,4
8306,0
60,0
075,0
026,5
225,0
0S
50
50,8
8S
50
50,8
8S
25
71,7
2-0
,00
337,5
075,0
00,0
027,5
21,7
5C
500
0,0
0G
200
0,0
0S
50
-0,0
0I11
10
138,7
4277,8
80,0
00,0
0-
0,0
6S
50
45,8
5S
50
45,8
5S
25
47,0
4123,4
2313,2
90,0
00,0
0-
0,1
9C
500
0,0
0G
200
55,9
7S
50
67,4
5I11
11
82,2
3254,3
190,1
90,0
0284,8
40,0
3S
50
27,1
7S
50
27,1
7S
25
27,8
8165,1
9337,5
00,0
090,1
9283,8
40,2
4C
500
0,0
0G
200
70,5
0S
50
94,6
9I12
00
97,5
0276,2
50,0
00,0
0-
0,0
6C
500
28,1
2G
100
50,6
3G
200
18,7
5150,0
0300,0
00,0
00,0
0-
8,5
7C
300
51,7
9G
100
62,5
0S
25
35,7
1I12
01
86,7
9276,2
510,7
10,0
0364,7
10,0
7C
500
21,8
2G
100
49,3
6G
200
15,6
0160,7
1300,0
00,0
010,7
1363,7
10,0
0C
300
62,5
0G
100
62,5
0S
25
35,7
1I12
10
97,5
0276,2
50,0
00,0
0-
0,0
6C
500
28,1
2G
100
50,6
3G
200
18,7
5150,0
0300,0
00,0
00,0
0-
0,0
2C
300
55,0
0G
100
60,0
0S
25
35,0
0I12
11
86,7
9276,2
510,7
10,0
088598,0
00,0
7C
500
21,8
2G
100
49,3
6G
200
15,6
0160,7
1300,0
00,0
010,7
188599,0
00,0
0C
300
62,5
0G
100
62,5
0S
25
35,7
1I13
00
135,0
0382,5
00,0
00,0
0-
0,7
2C
300
16,2
2G
200
81,2
7G
200
37,5
238,8
9315,0
00,0
00,0
0-
135,8
3C
500
19,4
4G
200
19,4
4G
100
0,0
0I13
01
173,8
9382,5
031,0
869,9
7231,7
40,6
3C
300
24,6
6G
200
95,0
3G
200
54,2
00,0
0315,0
069,9
731,0
8232,7
4165,0
0C
500
0,0
0G
200
0,0
0G
100
0,0
0I13
10
135,0
0382,5
00,0
00,0
0-
0,7
2C
300
16,2
2G
200
81,2
7G
200
37,5
283,9
5274,4
40,0
00,0
0-
0,2
5C
500
0,0
0G
200
49,7
8G
100
34,1
8I13
11
96,6
5382,5
071,2
732,9
2169,0
70,7
8C
300
0,0
0G
200
70,4
5G
200
26,1
9104,4
0290,5
632,9
271,2
7168,0
70,2
8C
500
0,0
0G
200
56,2
2G
100
48,1
8I14
00
90,0
0382,5
00,0
00,0
0-
2,8
7S
50
30,0
0S
50
30,0
0S
50
30,0
060,0
0255,0
00,0
00,0
0-
6,5
7C
500
37,2
5G
100
22,7
5S
50
0,0
0I14
01
150,0
0382,5
00,0
060,0
050,0
02,3
5S
50
50,0
0S
50
50,0
0S
50
50,0
00,0
0255,0
060,0
00,0
051,0
055,0
0C
500
0,0
0G
100
0,0
0S
50
0,0
0I14
10
207,6
1294,2
90,0
00,0
0-
0,0
9S
50
69,2
0S
50
69,2
0S
50
69,2
060,0
0255,0
00,0
00,0
0-
0,0
3C
500
30,9
9G
100
25,6
5S
50
3,3
6I14
11
233,0
2305,8
30,0
040,8
0290,6
60,1
0S
50
77,6
7S
50
77,6
7S
50
77,6
719,2
0255,0
040,8
00,0
0291,6
60,1
1C
500
0,0
0G
100
19,2
0S
50
0,0
0I15
00
60,0
0405,0
00,0
00,0
0-
118,1
6C
300
35,0
8S
50
12,4
6S
50
12,4
640,0
0268,0
00,0
00,0
0-
50,0
0S
50
13,3
3S
50
13,3
3S
50
13,3
3I15
01
0,0
0405,0
060,0
00,0
051,0
0155,0
0C
300
0,0
0S
50
0,0
0S
50
0,0
097,5
0270,0
00,0
060,0
050,0
026,1
3S
50
32,5
0S
50
32,5
0S
50
32,5
0
35
I15
10
128,7
9353,4
10,0
00,0
0-
0,3
4C
300
0,0
0S
50
64,3
9S
50
64,3
977,4
2238,0
60,0
00,0
0-
0,0
6S
50
25,8
1S
50
25,8
1S
50
25,8
1I15
11
201,6
4392,5
221,4
7146,4
760,8
80,4
8C
300
0,0
0S
50
100,8
2S
50
100,8
20,0
0200,0
0146,4
721,4
761,8
80,0
0S
50
0,0
0S
50
0,0
0S
50
0,0
0I16
00
60,0
0270,0
00,0
00,0
0-
0,1
0S
50
19,0
0S
50
19,0
0S
25
22,0
0128,5
7360,0
00,0
00,0
0-
110,0
0C
500
0,0
0G
100
34,1
3S
50
94,4
4I16
01
75,0
0270,0
00,0
015,0
099,0
00,0
0S
50
25,0
0S
50
25,0
0S
25
25,0
0113,5
7360,0
015,0
00,0
098,0
0110,0
0C
500
0,0
0G
100
19,1
3S
50
94,4
4I16
10
60,0
0270,0
00,0
00,0
0-
0,1
0S
50
19,0
0S
50
19,0
0S
25
22,0
0147,2
4346,9
30,0
00,0
0-
0,1
9C
500
0,4
1G
100
77,6
3S
50
69,2
0I16
11
17,7
5270,0
042,2
50,0
0543,0
00,3
7S
50
2,1
0S
50
2,1
0S
25
13,5
5170,8
3360,0
00,0
042,2
5544,0
00,2
2C
500
0,4
6G
100
88,0
9S
50
82,2
7I17
00
112,5
0318,7
50,0
00,0
0-
9,6
3C
500
48,7
5S
25
31,8
8S
25
31,8
843,7
5315,0
00,0
00,0
0-
0,3
8C
300
0,5
3G
200
28,7
5G
200
14,4
7I17
01
113,7
5318,7
531,6
832,9
3211,0
812,3
2C
500
44,9
0S
25
34,4
3S
25
34,4
342,5
0315,0
032,9
331,6
8212,0
80,3
8C
300
0,0
0G
200
28,4
0G
200
14,1
0I17
10
112,5
0318,7
50,0
00,0
0-
0,0
6C
500
21,4
6S
25
45,5
2S
25
45,5
243,7
5315,0
00,0
00,0
0-
0,3
8C
300
0,5
3G
200
28,7
5G
200
14,4
7I17
11
88,8
7318,7
523,6
30,0
038393,5
70,0
9C
500
0,2
9S
25
44,2
9S
25
44,2
967,3
8315,0
00,0
023,6
338394,5
70,3
4C
300
10,5
5G
200
35,4
3G
200
21,4
1I18
00
50,0
0337,5
00,0
00,0
0-
1,0
3G
100
0,0
0S
50
11,4
0S
25
38,6
0150,0
0360,0
00,0
00,0
0-
85,0
0C
300
0,0
0G
200
106,2
5G
200
43,7
5I18
01
200,0
0337,5
00,0
0150,0
073,5
60,0
9G
100
75,1
8S
50
61,2
8S
25
63,5
40,0
0360,0
0150,0
00,0
072,5
685,0
0C
300
0,0
0G
200
0,0
0G
200
0,0
0I18
10
84,1
0311,9
30,0
00,0
0-
0,3
0G
100
30,2
1S
50
22,9
7S
25
30,9
2150,0
0360,0
00,0
00,0
0-
0,1
8C
300
39,3
4G
200
61,6
4G
200
49,0
2I18
11
140,5
6337,5
00,0
090,5
66225,3
20,4
3G
100
42,6
8S
50
43,3
2S
25
54,5
679,9
1347,7
190,5
60,0
06226,3
20,2
1C
300
13,7
2G
200
39,4
4G
200
26,7
5I19
00
64,2
9405,0
00,0
00,0
0-
2,2
1S
50
21,4
3S
50
21,4
3S
50
21,4
360,0
0270,0
00,0
00,0
0-
0,2
6C
500
0,0
0G
200
16,5
7S
25
43,4
3I19
01
83,0
4405,0
00,0
018,7
550,0
02,1
0S
50
27,6
8S
50
27,6
8S
50
27,6
841,2
5270,0
018,7
50,0
051,0
00,3
5C
500
0,0
0G
200
0,0
0S
25
41,2
5I19
10
64,2
9405,0
00,0
00,0
0-
2,2
1S
50
21,4
3S
50
21,4
3S
50
21,4
378,7
4260,6
30,0
00,0
0-
0,1
1C
500
4,4
7G
200
36,4
3S
25
37,8
3I19
11
150,0
0405,0
016,6
1102,3
25570,4
81,7
0S
50
50,0
0S
50
50,0
0S
50
50,0
045,9
1234,1
9102,3
216,6
15569,4
80,0
6C
500
2,5
2G
200
20,5
4S
25
22,8
4I20
00
48,6
5331,2
20,0
00,0
0-
25,0
0S
25
16,2
2S
25
16,2
2S
50
16,2
2107,1
4300,0
00,0
00,0
0-
44,7
6C
500
29,3
7G
100
50,0
0S
50
27,7
8I20
01
142,5
0337,5
00,0
0100,8
349,0
00,0
0S
25
47,5
0S
25
47,5
0S
50
47,5
06,3
1300,0
0100,8
30,0
048,0
080,0
0C
500
0,0
0G
100
0,0
0S
50
6,3
1I20
10
68,1
4313,6
70,0
00,0
0-
0,0
3S
25
22,9
5S
25
22,9
5S
50
22,2
4107,1
4300,0
00,0
00,0
0-
0,0
7C
500
39,5
1G
100
43,4
8S
50
24,1
6I20
11
142,5
0337,5
00,0
0100,8
3499,2
40,0
0S
25
47,5
0S
25
47,5
0S
50
47,5
038,5
1277,4
6100,8
30,0
0500,2
40,1
1C
500
0,3
8G
100
24,5
1S
50
13,6
2I21
00
43,7
5315,0
00,0
00,0
0-
0,9
8S
50
14,5
8S
50
14,5
8S
50
14,5
878,5
7220,0
00,0
00,0
0-
3,0
7C
500
51,4
8S
50
13,5
4S
50
13,5
4I21
01
122,3
2315,0
00,0
078,5
750,0
00,4
0S
50
40,7
7S
50
40,7
7S
50
40,7
70,0
0220,0
078,5
70,0
051,0
070,0
0C
500
0,0
0S
50
0,0
0S
50
0,0
0I21
10
84,2
1282,6
30,0
00,0
0-
0,0
4S
50
28,0
7S
50
28,0
7S
50
28,0
778,5
7220,0
00,0
00,0
0-
0,0
2C
500
47,8
2S
50
15,3
7S
50
15,3
7I21
11
177,2
7315,0
00,0
0133,5
2583,6
30,0
0S
50
59,0
9S
50
59,0
9S
50
59,0
97,8
6176,0
3133,5
20,0
0584,6
30,0
5C
500
0,1
0S
50
3,8
8S
50
3,8
8I22
00
36,6
7247,5
00,0
00,0
0-
13,9
4C
300
25,8
2G
200
10,8
5S
50
0,0
062,5
0337,5
00,0
00,0
0-
2,0
8C
500
0,0
0S
50
12,5
0S
25
50,0
0I22
01
0,0
0247,5
036,6
70,0
051,0
047,5
0C
300
0,0
0G
200
0,0
0S
50
0,0
099,1
7337,5
00,0
036,6
750,0
01,2
7C
500
0,0
0S
50
49,1
7S
25
50,0
0I22
10
36,6
7247,5
00,0
00,0
0-
0,0
6C
300
23,6
2G
200
13,0
4S
50
0,0
0100,4
6314,7
30,0
00,0
0-
0,2
3C
500
0,4
6S
50
50,0
0S
25
50,0
0I22
11
0,3
2246,2
238,0
50,0
0777,9
10,1
5C
300
0,3
2G
200
0,0
0S
50
0,0
0100,5
5337,5
00,0
038,0
5776,9
10,2
7C
500
0,5
5S
50
50,0
0S
25
50,0
0I23
00
100,0
0300,0
00,0
00,0
0-
38,3
3S
25
33,3
3S
25
33,3
3S
25
33,3
370,0
0315,0
00,0
00,0
0-
19,0
0G
100
10,0
0S
50
30,0
0S
50
30,0
0I23
01
170,0
0300,0
00,0
070,0
025,0
09,1
7S
25
56,6
7S
25
56,6
7S
25
56,6
70,0
0315,0
070,0
00,0
024,0
025,0
0G
100
0,0
0S
50
0,0
0S
50
0,0
0I23
10
100,0
0300,0
00,0
00,0
0-
1,5
3S
25
33,3
3S
25
33,3
3S
25
33,3
370,0
0315,0
00,0
00,0
0-
0,0
9G
100
27,2
7S
50
21,3
6S
50
21,3
6I23
11
158,9
1300,0
00,0
058,9
1441,6
70,5
5S
25
52,9
7S
25
52,9
7S
25
52,9
711,0
9315,0
058,9
10,0
0442,6
70,2
5G
100
0,4
9S
50
5,3
0S
50
5,3
0I24
00
120,0
0360,0
00,0
00,0
0-
0,7
3C
500
0,0
0G
100
66,6
7G
100
53,3
370,0
0297,5
00,0
00,0
0-
38,8
8C
500
30,9
0G
100
30,9
0G
100
8,2
1I24
01
184,8
6360,0
02,8
567,7
1227,3
10,3
2C
500
0,0
0G
100
98,4
6G
100
86,4
05,1
4297,5
067,7
12,8
5226,3
174,9
7C
500
1,6
3G
100
2,0
3G
100
1,4
8I24
10
153,4
1334,9
40,0
00,0
0-
0,2
7C
500
0,0
0G
100
83,0
4G
100
70,3
674,9
1293,8
20,0
00,0
0-
0,1
5C
500
0,0
0G
100
47,2
4G
100
27,6
7I24
11
148,5
2331,0
510,0
80,0
085,1
20,2
6C
500
0,0
0G
100
80,6
5G
100
67,8
780,0
8297,5
00,0
010,0
886,1
20,1
6C
500
0,0
0G
100
49,6
0G
100
30,4
8I25
00
70,0
0315,0
00,0
00,0
0-
95,0
0C
500
0,0
0G
200
0,0
0S
50
70,0
042,8
6270,0
00,0
00,0
0-
39,1
1G
200
24,7
1S
25
9,0
7S
25
9,0
7I25
01
81,2
5315,0
021,4
532,7
090,9
895,0
0C
500
0,0
0G
200
0,0
0S
50
81,2
531,6
1270,0
032,7
021,4
589,9
853,4
1G
200
11,8
5S
25
9,8
8S
25
9,8
8I25
10
70,0
0315,0
00,0
00,0
0-
0,4
0C
500
0,0
0G
200
14,0
5S
50
55,9
555,6
7261,0
30,0
00,0
0-
0,3
1G
200
0,0
0S
25
27,8
4S
25
27,8
4I25
11
49,9
7312,0
725,8
90,0
057,8
30,4
6C
500
0,0
0G
200
0,0
0S
50
49,9
768,7
5270,0
00,0
025,8
958,8
30,3
5G
200
0,0
0S
25
34,3
8S
25
34,3
8I26
00
135,0
0382,5
00,0
00,0
0-
0,7
4C
500
0,0
0G
200
15,3
6S
50
119,6
490,0
0382,5
00,0
00,0
0-
70,7
1C
500
38,7
3S
25
25,6
4S
50
25,6
4I26
01
114,8
4382,5
073,1
152,9
6124,2
30,8
1C
500
0,0
0G
200
0,0
0S
50
114,8
4110,1
6382,5
052,9
673,1
1123,2
363,2
6C
500
20,5
8S
25
44,7
9S
50
44,7
9I26
10
192,8
9353,5
60,0
00,0
0-
0,2
7C
500
0,0
0G
200
80,1
3S
50
112,7
5131,2
8351,5
40,0
00,0
0-
0,1
5C
500
0,0
0S
25
67,8
9S
50
63,3
9I26
11
150,2
4320,5
4175,3
866,7
1102,7
60,2
0C
500
0,0
0G
200
60,3
3S
50
89,9
1198,6
8382,5
066,7
1175,3
8103,7
60,2
2C
500
0,0
0S
25
102,5
0S
50
96,1
8I27
00
115,2
2305,8
70,0
00,0
0-
50,0
0S
50
32,6
1S
50
32,6
1S
25
50,0
075,0
0337,5
00,0
00,0
0-
82,8
8C
300
34,6
2G
200
34,6
2G
200
5,7
7I27
01
142,8
6334,2
98,3
383,3
322,4
82,3
8S
50
47,6
2S
50
47,6
2S
25
47,6
20,0
0337,5
083,3
38,3
323,4
8117,5
0C
300
0,0
0G
200
0,0
0G
200
0,0
0I27
10
62,5
0337,5
00,0
00,0
0-
2,2
1S
50
6,2
5S
50
6,2
5S
25
50,0
093,1
2328,4
40,0
00,0
0-
0,3
0C
300
19,7
2G
200
49,2
9G
200
24,1
1I27
11
150,0
0337,5
03,5
391,0
35779,9
81,2
5S
50
50,0
0S
50
50,0
0S
25
50,0
067,1
0297,7
091,0
33,5
35778,9
80,2
1C
300
14,1
3G
200
35,3
2G
200
17,6
6I28
00
80,0
0240,0
00,0
00,0
0-
14,4
7C
300
55,9
4G
200
19,6
3S
25
4,4
2150,0
0360,0
00,0
00,0
0-
0,4
7G
200
0,0
0S
25
83,3
3S
50
66,6
7I28
01
0,0
0240,0
080,0
00,0
0153,5
090,0
0C
300
0,0
0G
200
0,0
0S
25
0,0
0230,0
0360,0
00,0
080,0
0152,5
00,2
1G
200
52,8
8S
25
92,3
8S
50
84,7
5I28
10
80,0
0240,0
00,0
00,0
0-
0,0
8C
300
47,9
1G
200
17,7
8S
25
14,3
1152,0
8358,7
50,0
00,0
0-
0,4
2G
200
0,3
5S
25
83,3
1S
50
68,4
1I28
11
26,7
7240,0
053,2
30,0
028153,5
00,2
3C
300
15,3
8G
200
0,0
0S
25
11,3
8203,2
3360,0
00,0
053,2
328152,5
00,2
9G
200
34,1
4S
25
89,7
0S
50
79,4
0I29
00
75,0
0382,5
00,0
00,0
0-
4,4
0S
25
37,5
0S
50
0,0
0S
25
37,5
040,6
2292,5
00,0
00,0
0-
0,1
6C
500
0,0
0G
200
31,1
2G
200
9,5
1I29
01
115,6
3382,5
00,0
040,6
225,0
03,2
6S
25
57,8
1S
50
0,0
0S
25
57,8
10,0
0292,5
040,6
20,0
026,0
00,3
6C
500
0,0
0G
200
0,0
0G
200
0,0
0I29
10
75,0
0382,5
00,0
00,0
0-
4,4
0S
25
37,5
0S
50
0,0
0S
25
37,5
040,6
2292,5
00,0
00,0
0-
0,1
6C
500
0,0
0G
200
31,1
2G
200
9,5
1I29
11
69,8
6382,5
033,8
828,7
5425,8
84,8
9S
25
28,8
3S
50
12,2
1S
25
28,8
345,7
6292,5
028,7
533,8
8426,8
80,1
4C
500
0,0
1G
200
33,4
7G
200
12,2
8I30
00
112,5
0360,0
00,0
00,0
0-
54,1
4C
300
50,7
8S
50
30,8
6S
50
30,8
636,1
1292,5
00,0
00,0
0-
0,0
0S
50
15,9
1S
50
4,2
9S
25
15,9
1I30
01
100,8
8360,0
070,4
758,8
5141,7
668,9
0C
300
37,3
6S
50
31,7
6S
50
31,7
647,7
3292,5
058,8
570,4
7142,7
60,0
0S
50
15,9
1S
50
15,9
1S
25
15,9
1I30
10
121,5
1352,7
90,0
00,0
0-
0,3
4C
300
0,1
4S
50
60,6
9S
50
60,6
936,1
1292,5
00,0
00,0
0-
0,1
0S
50
11,2
6S
50
11,2
6S
25
13,5
9I30
11
133,5
3360,0
00,0
021,0
3227,1
50,3
7C
300
0,1
5S
50
66,6
9S
50
66,6
920,0
9287,9
921,0
30,0
0226,1
50,1
4S
50
5,6
7S
50
5,6
7S
25
8,7
4I31
00
46,6
7315,0
00,0
00,0
0-
56,6
2C
300
30,7
0S
25
7,9
8S
25
7,9
875,0
0382,5
00,0
00,0
0-
162,5
0C
300
0,0
0G
100
0,0
0S
50
75,0
0I31
01
-0,0
0315,0
046,6
70,0
0101,0
095,0
0C
300
-0,0
0S
25
0,0
0S
25
0,0
0121,6
7382,5
00,0
046,6
7100,0
0162,5
0C
300
0,0
0G
100
23,9
4S
50
97,7
3I31
10
73,5
9294,8
10,0
00,0
0-
0,2
5C
300
0,1
0S
25
36,7
5S
25
36,7
5104,2
7356,1
50,0
00,0
0-
0,4
5C
300
0,3
0G
100
50,7
6S
50
53,2
1I31
11
108,8
8315,0
038,7
0100,9
2133,0
70,3
2C
300
0,1
3S
25
54,3
8S
25
54,3
884,0
9318,3
3100,9
238,7
0132,0
70,3
3C
300
0,2
2G
100
39,5
8S
50
44,2
8I32
00
86,6
7260,0
00,0
00,0
0-
0,0
8C
300
27,5
1S
50
28,7
7S
25
30,3
865,0
0276,2
50,0
00,0
0-
3,5
0S
25
21,6
7S
50
21,6
7S
50
21,6
7I32
01
76,6
7260,0
010,0
00,0
0246,6
90,1
2C
300
19,5
8S
50
27,3
9S
25
29,7
075,0
0276,2
50,0
010,0
0245,6
90,0
0S
25
25,0
0S
50
25,0
0S
50
25,0
0I32
10
86,6
7260,0
00,0
00,0
0-
0,0
8C
300
27,5
1S
50
28,7
7S
25
30,3
868,1
8273,8
60,0
00,0
0-
-0,0
0S
25
22,7
3S
50
22,7
3S
50
22,7
3I32
11
76,6
7260,0
010,0
00,0
071673,2
60,1
2C
300
19,5
8S
50
27,3
9S
25
29,7
075,0
0276,2
50,0
010,0
071674,2
60,0
0S
25
25,0
0S
50
25,0
0S
50
25,0
0I33
00
75,0
0318,7
50,0
00,0
0-
0,0
8C
300
21,0
8G
200
29,4
0G
100
24,5
254,1
7292,5
00,0
00,0
0-
0,0
9C
300
17,0
1G
100
37,1
5G
200
0,0
0I33
01
97,6
9318,7
50,0
022,6
9230,7
70,0
4C
300
32,2
0G
200
36,8
2G
100
28,6
631,4
8292,5
022,6
90,0
0231,7
70,1
4C
300
0,0
0G
100
31,4
8G
200
0,0
0I33
10
75,0
0318,7
50,0
00,0
0-
0,0
8C
300
21,0
8G
200
29,4
0G
100
24,5
254,1
7292,5
00,0
00,0
0-
0,0
9C
300
17,0
1G
100
37,1
5G
200
0,0
0I33
11
97,0
6318,7
50,0
022,0
644365,3
80,0
4C
300
31,9
0G
200
36,6
2G
100
28,5
532,1
0292,5
022,0
60,0
044366,3
80,1
4C
300
0,4
7G
100
31,6
4G
200
0,0
0I34
00
69,6
4276,2
50,0
00,0
0-
0,0
8C
300
26,4
6G
200
33,2
6G
200
9,9
297,5
0276,2
50,0
00,0
0-
0,2
5S
25
32,5
0S
50
32,5
0S
25
32,5
0I34
01
68,7
1276,2
526,5
525,6
1243,8
80,0
9C
300
23,8
4G
200
31,5
2G
200
13,3
598,4
4276,2
525,6
126,5
5244,8
80,0
0S
25
32,8
1S
50
32,8
1S
25
32,8
1I34
10
69,6
4276,2
50,0
00,0
0-
0,0
8C
300
26,4
6G
200
33,2
6G
200
9,9
297,5
0276,2
50,0
00,0
0-
0,0
1S
25
32,5
8S
50
32,3
4S
25
32,5
8
36
I34
11
68,7
1276,2
526,5
225,5
867586,1
00,0
9C
300
23,8
5G
200
31,5
2G
200
13,3
498,4
4276,2
525,5
826,5
267587,1
00,0
0S
25
32,8
1S
50
32,8
1S
25
32,8
1I35
00
107,1
4300,0
00,0
00,0
0-
0,0
3C
500
37,4
5G
100
47,4
9G
200
22,2
0100,0
0360,0
00,0
00,0
0-
0,6
6G
100
3,5
7S
25
48,2
1S
25
48,2
1I35
01
37,5
0300,0
069,6
40,0
0101,0
00,1
3C
500
0,0
0G
100
37,5
0G
200
0,0
0169,6
4360,0
00,0
069,6
4100,0
00,0
0G
100
69,6
4S
25
50,0
0S
25
50,0
0I35
10
107,1
4300,0
00,0
00,0
0-
0,0
3C
500
37,4
5G
100
47,4
9G
200
22,2
0119,5
3342,4
30,0
00,0
0-
0,1
7G
100
34,9
5S
25
42,2
9S
25
42,2
9I35
11
45,5
7300,0
061,5
70,0
010101,0
00,1
0C
500
0,0
1G
100
40,0
0G
200
5,5
6161,5
7360,0
00,0
061,5
710100,0
00,0
8G
100
61,5
7S
25
50,0
0S
25
50,0
0I36
00
75,0
0337,5
00,0
00,0
0-
59,3
2C
500
35,2
3G
100
35,2
3G
100
4,5
570,0
0315,0
00,0
00,0
0-
0,0
0S
25
41,6
7S
50
0,0
0S
50
28,3
3I36
01
20,0
0337,5
069,9
814,9
8233,4
381,6
4C
500
6,7
0G
100
7,3
2G
100
5,9
8125,0
0315,0
014,9
869,9
8234,4
30,0
0S
25
41,6
7S
50
41,6
7S
50
41,6
7I36
10
112,0
7318,9
70,0
00,0
0-
0,1
4C
500
0,0
0G
100
68,9
7G
100
43,1
085,7
1307,1
40,0
00,0
0-
0,0
0S
25
28,5
7S
50
28,5
7S
50
28,5
7I36
11
151,7
9337,5
00,0
076,7
9277,3
30,1
8C
500
0,0
0G
100
87,5
0G
100
64,2
99,4
3306,8
976,7
90,0
0278,3
30,4
5S
25
9,4
3S
50
0,0
0S
50
0,0
0I37
00
68,7
5220,0
00,0
00,0
0-
64,4
9C
300
3,9
4G
100
50,0
0G
200
14,8
1135,0
0382,5
00,0
00,0
0-
98,5
0G
100
50,0
0S
25
42,5
0S
25
42,5
0I37
01
0,0
0220,0
0131,6
062,8
535,1
370,0
0C
300
0,0
0G
100
0,0
0G
200
0,0
0203,7
5382,5
062,8
5131,6
034,1
375,6
6G
100
26,7
2S
25
88,5
1S
25
88,5
1I37
10
68,7
5220,0
00,0
00,0
0-
0,1
1C
300
26,5
6G
100
42,1
9G
200
0,0
0183,8
7358,0
60,0
00,0
0-
1,3
8G
100
0,0
0S
25
91,9
4S
25
91,9
4I37
11
16,8
0185,2
895,3
50,0
085,5
20,1
2C
300
0,0
0G
100
16,8
0G
200
0,0
0230,3
5382,5
00,0
095,3
584,5
21,6
2G
100
0,4
1S
25
114,9
7S
25
114,9
7I38
00
50,0
0337,5
00,0
00,0
0-
128,9
2C
300
43,4
0G
100
6,6
0S
50
0,0
050,0
0405,0
00,0
00,0
0-
99,2
8C
300
27,9
3G
200
15,7
2S
50
6,3
5I38
01
100,0
0337,5
034,8
484,8
4164,2
9130,6
4C
300
39,4
2G
100
32,0
7S
50
28,5
00,0
0405,0
084,8
434,8
4165,2
9130,0
0C
300
0,0
0G
200
0,0
0S
50
0,0
0I38
10
90,6
9306,9
80,0
00,0
0-
0,5
2C
300
0,0
0G
100
42,0
4S
50
48,6
571,7
9385,3
80,0
00,0
0-
0,3
7C
300
0,0
0G
200
21,7
9S
50
50,0
0I38
11
126,6
9337,5
00,0
076,6
988,8
40,6
3C
300
0,0
0G
100
57,6
9S
50
69,0
042,5
0342,7
276,6
90,0
089,8
40,2
3C
300
0,0
0G
200
7,5
7S
50
34,9
3I39
00
50,0
0337,5
00,0
00,0
0-
128,9
2C
300
43,4
0G
100
6,6
0S
50
0,0
069,6
4276,2
50,0
00,0
0-
47,2
3S
50
23,2
1S
50
23,2
1S
50
23,2
1I39
01
0,0
0337,5
050,0
00,0
051,0
0187,5
0C
300
0,0
0G
100
0,0
0S
50
0,0
0119,6
4276,2
50,0
050,0
050,0
026,4
0S
50
39,8
8S
50
39,8
8S
50
39,8
8I39
10
90,6
9306,9
80,0
00,0
0-
0,5
2C
300
0,0
0G
100
42,0
4S
50
48,6
569,6
4276,2
50,0
00,0
0-
0,9
4S
50
23,2
1S
50
23,2
1S
50
23,2
1I39
11
120,5
6328,4
314,5
172,9
887,5
90,5
9C
300
0,0
0G
100
54,9
0S
50
65,6
611,1
7276,2
572,9
814,5
188,5
91,4
3S
50
3,7
2S
50
3,7
2S
50
3,7
2I40
00
72,2
2260,0
00,0
00,0
0-
38,1
3G
200
15,0
8S
25
28,5
7S
50
28,5
760,0
0255,0
00,0
00,0
0-
0,8
7S
25
25,5
8S
25
25,5
8S
50
8,8
5I40
01
28,5
7260,0
043,6
50,0
036,5
060,0
0G
200
0,0
0S
25
28,5
7S
50
-0,0
0103,6
5255,0
00,0
043,6
537,5
00,3
0S
25
36,4
9S
25
36,4
9S
50
30,6
7I40
10
73,3
4259,0
00,0
00,0
0-
0,1
2G
200
24,0
8S
25
25,7
0S
50
23,5
560,0
0255,0
00,0
00,0
0-
0,8
7S
25
25,5
8S
25
25,5
8S
50
8,8
5I40
11
38,1
2230,4
668,3
71,4
53501,3
90,0
6G
200
12,4
3S
25
13,4
0S
50
12,2
9126,9
2255,0
01,4
568,3
73502,3
90,0
0S
25
42,3
1S
25
42,3
1S
50
42,3
1I41
00
112,5
0318,7
50,0
00,0
0-
0,5
5C
300
2,1
0G
100
102,9
7G
200
7,4
2120,0
0360,0
00,0
00,0
0-
85,0
0G
200
0,0
0S
50
82,3
5S
50
37,6
5I41
01
67,7
9318,7
544,7
10,0
098,0
00,9
4C
300
0,0
0G
100
67,7
9G
200
0,0
0164,7
1360,0
00,0
044,7
199,0
085,0
0G
200
0,0
0S
50
82,3
5S
50
82,3
5I41
10
112,5
0318,7
50,0
00,0
0-
0,5
5C
300
2,1
0G
100
102,9
7G
200
7,4
2120,4
3359,6
80,0
00,0
0-
0,4
0G
200
4,0
4S
50
58,2
0S
50
58,2
0I41
11
108,6
1318,7
53,8
90,0
012977,4
20,5
6C
300
0,1
9G
100
102,3
4G
200
6,0
8123,8
9360,0
00,0
03,8
912976,4
20,4
0G
200
5,8
9S
50
59,0
0S
50
59,0
0I42
00
40,0
0270,0
00,0
00,0
0-
23,8
1C
500
20,9
5G
100
19,0
5S
25
0,0
093,7
5318,7
50,0
00,0
0-
0,5
6S
50
19,5
3S
25
37,1
1S
25
37,1
1I42
01
0,0
0270,0
040,0
00,0
038,5
050,0
0C
500
0,0
0G
100
0,0
0S
25
0,0
0133,7
5318,7
50,0
040,0
037,5
00,2
4S
50
39,5
3S
25
47,1
1S
25
47,1
1I42
10
40,0
0270,0
00,0
00,0
0-
0,0
4C
500
24,6
2G
100
15,3
8S
25
0,0
093,7
5318,7
50,0
00,0
0-
0,5
6S
50
19,5
3S
25
37,1
1S
25
37,1
1I42
11
9,6
1270,0
030,3
90,0
01288,5
00,1
0C
500
0,0
7G
100
9,5
4S
25
0,0
0124,1
4318,7
50,0
030,3
91287,5
00,3
2S
50
34,7
3S
25
44,7
1S
25
44,7
1I43
00
55,0
0247,5
00,0
00,0
0-
0,0
0C
300
27,5
0G
200
27,5
0G
200
0,0
087,5
0297,5
00,0
00,0
0-
1,0
9C
300
51,5
7G
100
26,7
7S
25
9,1
6I43
01
50,8
8247,5
04,1
20,0
0258,0
00,0
1C
300
25,0
3G
200
25,8
5G
200
0,0
091,6
2297,5
00,0
04,1
2259,0
00,0
0C
300
52,7
8G
100
28,1
3S
25
10,7
1I43
10
55,0
0247,5
00,0
00,0
0-
0,0
0C
300
27,5
0G
200
27,5
0G
200
0,0
087,5
0297,5
00,0
00,0
0-
0,0
1C
300
50,0
0G
100
27,0
8S
25
10,4
2I43
11
50,8
8247,5
04,1
20,0
090258,0
00,0
1C
300
25,0
3G
200
25,8
5G
200
0,0
091,6
2297,5
00,0
04,1
290259,0
00,0
0C
300
52,7
8G
100
28,1
3S
25
10,7
1I44
00
125,0
0300,0
00,0
00,0
0-
64,7
2G
200
13,8
9S
25
55,5
6S
50
55,5
666,6
7240,0
00,0
00,0
0-
0,0
0S
25
27,5
9S
25
27,5
9S
25
11,4
9I44
01
108,9
1300,0
016,0
90,0
050,0
080,0
0G
200
0,0
0S
25
55,5
6S
50
53,3
582,7
6240,0
00,0
016,0
949,0
00,0
0S
25
27,5
9S
25
27,5
9S
25
27,5
9I44
10
125,0
0300,0
00,0
00,0
0-
0,2
2G
200
32,3
8S
25
49,3
9S
50
43,2
366,6
7240,0
00,0
00,0
0-
0,3
1S
25
22,2
2S
25
22,2
2S
25
22,2
2I44
11
108,9
1300,0
016,0
90,0
013864,5
20,2
8G
200
21,3
5S
25
47,7
1S
50
39,8
682,7
6240,0
00,0
016,0
913863,5
20,0
0S
25
27,5
9S
25
27,5
9S
25
27,5
9I45
00
128,5
7360,0
00,0
00,0
0-
0,4
0C
300
16,0
0G
100
83,1
1G
200
29,4
650,0
0315,0
00,0
00,0
0-
40,0
0C
300
0,0
0G
200
44,4
4G
200
5,5
6I45
01
102,8
8360,0
025,6
90,0
0229,2
50,4
5C
300
3,9
5G
100
79,1
0G
200
19,8
375,6
9315,0
00,0
025,6
9230,2
540,0
0C
300
-0,0
0G
200
44,4
4G
200
31,2
5I45
10
128,5
7360,0
00,0
00,0
0-
0,4
0C
300
16,0
0G
100
83,1
1G
200
29,4
650,0
0315,0
00,0
00,0
0-
0,0
9C
300
15,4
2G
200
25,1
0G
200
9,4
8I45
11
96,5
2360,0
044,7
712,7
138214,9
10,4
6C
300
0,9
7G
100
78,1
0G
200
17,4
482,0
5315,0
012,7
144,7
738213,9
10,0
5C
300
28,3
8G
200
33,7
3G
200
19,9
5I46
00
43,3
3292,5
00,0
00,0
0-
0,0
9C
300
15,6
4S
50
13,8
5S
50
13,8
568,7
5233,7
50,0
00,0
0-
0,9
2C
500
28,5
4G
200
28,5
4G
100
11,6
6I46
01
77,3
2292,5
01,1
935,1
8305,1
80,0
0C
300
40,4
8S
50
18,4
2S
50
18,4
234,7
7233,7
535,1
81,1
9304,1
817,2
7C
500
14,3
3G
200
14,3
3G
100
6,1
0I46
10
43,3
3292,5
00,0
00,0
0-
0,0
9C
300
15,6
4S
50
13,8
5S
50
13,8
568,7
5233,7
50,0
00,0
0-
0,0
0C
500
27,8
3G
200
28,7
4G
100
12,1
8I46
11
77,3
2292,5
00,0
033,9
8155739,6
80,0
0C
300
40,4
8S
50
18,4
2S
50
18,4
234,7
7233,7
533,9
80,0
0155740,6
80,0
5C
500
6,7
1G
200
20,2
9G
100
7,7
7I47
00
87,5
0297,5
00,0
00,0
0-
77,5
0C
500
0,0
0G
200
34,7
2G
200
52,7
875,0
0337,5
00,0
00,0
0-
0,7
8S
50
14,1
7S
25
46,6
7S
50
14,1
7I47
01
-0,0
0297,5
087,5
00,0
046,0
077,5
0C
500
0,0
0G
200
-0,0
0G
200
0,0
0162,5
0337,5
00,0
087,5
045,0
00,3
6S
50
49,1
7S
25
64,1
7S
50
49,1
7I47
10
87,5
0297,5
00,0
00,0
0-
0,1
2C
500
14,2
4G
200
47,9
7G
200
25,2
989,6
1330,1
90,0
00,0
0-
0,5
3S
50
22,4
5S
25
44,7
2S
50
22,4
5I47
11
19,2
8240,9
3162,5
00,0
07183,8
00,0
4C
500
1,5
4G
200
12,0
3G
200
5,7
0237,5
0337,5
00,0
0162,5
07182,8
00,0
0S
50
79,1
7S
25
79,1
7S
50
79,1
7I48
00
54,2
2276,2
00,0
00,0
0-
50,0
0S
25
18,0
7S
25
18,0
7S
50
18,0
739,2
9247,5
00,0
00,0
0-
24,5
1C
300
18,4
0G
100
18,4
0S
50
2,5
0I48
01
93,4
5276,2
50,0
039,2
925,0
043,2
4S
25
22,7
6S
25
52,5
9S
50
18,1
00,0
0247,5
039,2
90,0
026,0
047,5
0C
300
0,0
0G
100
0,0
0S
50
0,0
0I48
10
54,1
7276,2
50,0
00,0
0-
1,5
0S
25
26,6
9S
25
26,6
9S
50
0,7
940,6
4246,5
50,0
00,0
0-
0,1
5C
300
0,1
0G
100
24,8
6S
50
15,6
8I48
11
51,6
4276,2
531,2
528,7
2434,5
21,9
8S
25
18,5
3S
25
18,5
3S
50
14,5
841,8
1247,5
028,7
231,2
5435,5
20,1
6C
300
0,1
1G
100
25,3
7S
50
16,3
4I49
00
69,6
4276,2
50,0
00,0
0-
0,0
8C
300
26,4
6G
200
33,2
6G
200
9,9
270,3
1318,7
50,0
00,0
0-
0,6
0C
300
0,0
0S
50
35,1
6S
50
35,1
6I49
01
15,6
3276,2
554,0
20,0
0230,3
50,1
9C
300
0,0
0G
200
15,6
3G
200
0,0
0124,3
3318,7
50,0
054,0
2229,3
50,2
4C
300
20,9
9S
50
51,6
7S
50
51,6
7I49
10
69,6
4276,2
50,0
00,0
0-
0,0
8C
300
26,4
6G
200
33,2
6G
200
9,9
290,4
8302,6
10,0
00,0
0-
0,2
1C
300
14,2
4S
50
38,1
2S
50
38,1
2I49
11
15,7
3276,2
553,9
10,0
064793,5
70,1
9C
300
0,0
6G
200
15,6
7G
200
0,0
0124,2
2318,7
50,0
053,9
164794,5
70,2
4C
300
20,9
1S
50
51,6
6S
50
51,6
6
37