Concentration of ground states in stationary Mean-Field Games systems: part I

Size: px
Start display at page:

Download "Concentration of ground states in stationary Mean-Field Games systems: part I"

Transcription

1 Concentration of ground states in stationary Mean-Field Gaes systes: part I Annalisa Cesaroni and Marco Cirant June 9, 207 Abstract In this paper we provide the existence of classical solutions to soe stationary ean field gae systes in the whole space, with coercive potential and aggregating local coupling. This result is obtained under structural assuptions on the growth at infinity of the coupling ter and of the Hailtonian, by using a variational approach based on the analysis of the non-convex energy associated to the syste. Moreover, we analyse the liit of the syste as the diffusion vanishes, and we show that phenoena of concentration of ass appear. We also describe the asyptotic shape of the rescaled solutions in the vanishing viscosity liit, in particular proving the existence of ground states, i.e. classical solutions to ean field gae systes in the whole space without potential, and with aggregating coupling. AMS-Subject Classification. 35J50, 4970, 35J47, 9A3, 35B25 Keywords. Ergodic Mean-Field Gaes, Concentration-copactness ethod, Mass concentration, Seiclassical liit, Elliptic systes, Variational ethods. Contents Introduction 2 Soe preliinary regularity results 6 2. The Hailton-Jacobi-Bellan equation on the whole space A-priori estiates for the Kologorov equation Existence of solutions to the MFG syste for > A-priori estiates and energy bounds Existence of iniizers of E and solutions to for every > Convergence to ground states and concentration The rescaled proble A preliinary convergence result Concentration-copactness Convergence to ground states References 33 Introduction This paper is devoted to the analysis of ergodic Mean-Field Gaes systes with local decreasing coupling. The proble we consider is the following: given, M > 0, find a constant λ R for

2 which there exists a couple u, solving u + H u + λ = α + V x div H u = 0 on R, = M, where H : is a convex function, V : [0, + is a coercive potential, and α > 0. The second proble we address is the analysis of the asyptotic behaviour of the solutions as 0. Mean-Field Gaes MFG is a recent theory that odels the behaviour of a very large nuber of indistinguishable rational agents aiing at iniizing a coon cost. The theory was introduced in the seinal papers by Lasry, Lions [23, 24, 25, 26] and by Huang, Caines, Malhaé [9], and has been rapidly growing during the last decade due to its atheatical challenges and several potential applications fro econoics and finance, to engineering and odels of social systes. In the ergodic MFG setting, the dynaics of a typical agent is given by the controlled stochastic differential equation dx s = v s ds + 2 db s, where v s is the control and B s is a Brownian otion, and the cost of long-tie average for is given by T li T T E [Lv s + V X s α X s ]ds, 0 where the Lagrangian L is the Legendre transfor of H see 5 and x denotes the density of population of sall agents at a position x. In an equilibriu regie, a typical agent iniizes his own cost, his corresponding density is stable as tie s, and coincides with the population density. This equilibriu is encoded fro the PDE viewpoint in : a solution u of the Hailton-Jacobi-Bellan equation gives an optial control for the typical agent in feedback for H u, and the Kologorov equation provides the density of the agent playing in an optial way. The two key points of our setting are the following: firstly, the cost is onotonically decreasing with respect to the population distribution, naely agents are attracted toward congested areas. A large part of the MFG literature focuses on the study of systes with copetition, naely when the coupling in the cost is onotonically increasing; this assuption is essential if one seeks for uniqueness of equilibria, and it is in general crucial in any existence and regularity arguents, see, e.g [6], and references therein. On the other hand, odels with aggregation like have been considered in few cases, see [3, 4, 5]. Secondly, the state of a typical agent here is the whole euclidean space. Usually, the analysis of is carried out in the periodic setting, in order to avoid boundary issues and the non-copactness of. Few investigations are available in the truly non-periodic setting: see [30] for tie-dependent probles, [2] for the case of bounded controls, [7] for soe regularity results and [3] for the Linear-Quadratic fraework. We observe that the non-copact setting is even ore delicate for stationary ergodic probles like : a stable long-tie regie of a typical player is ensured if the Brownian otion is copensated by the optial velocity v s. In other words, if a force that drives players to bounded states is issing, dissipation eventually leads their distribution to vanish on the whole. This phenoenon is ipossible if the state space is copact. The ain issue here is that the behaviour of the optial velocity v s = H u is a-priori unknown, and depends in an iplicit way on V and the distribution itself. Let us ention that even without the coupling α, the ergodic control proble in unbounded doains has received a considerable attention, see e.g. [5, 20, 2] and references therein. The first assuption we ake is then the following: V C 0,η loc R for soe η 0, ], and C V ax{ x C V, 0} b V x C V + x b 2 2

3 for soe C V > 0 and b > 0. ote that V represents the spatial preference of a single agent; if it grows as x, it discourages agents to be far away fro the origin. At the PDE level, this will copensate the lack of copactness of. The requireent of V to be positive is just technical. Before stating the assuptions on H and α, let us coent on the viewpoint we will rely on. Since the early work [24], it is known that is associated to a control proble of PDE see also [28]. Let K,M be the set of, w L W,γ L γ satisfying ϕ dx = w ϕ dx ϕ C0, R 3 dx = M, 0 a.e.. We define the energy E, w := L w dx + V x dx α+ dx, 4 R R α + L is the Legendre transfor of H, i.e. Lq = H q = sup p [p q Hp], q. 5 ote that the expression L w/ has to be intended equal to zero if, w = 0, 0 and + if, w 0, 0 and 0: L / thus reads as the Legendre transfor of H. For any > 0, M > 0 fixed, we will look for solutions of as iniizers of E, that is we will consider the iniization proble e M = inf,w K,M E, w. 6 Once a iniizer, w is available, one can indeed construct a triple u, λ, solving. To coplete this progra, we will assue that H C 2 \ {0} is strictly convex, and there exist soe C H > 0, and γ > such that C H p γ C H Hp C H p γ p. 7 ote that fro 7 and 5, there exists C L > 0 such that C L q γ Lq C L q γ + C L, where = γ γ. We will also always assue that the power growth α of the coupling and the growth of the Lagrangian satisfy 0 < α < γ. 8 Moreover, we will require γ <, i.e. γ >. 9 Soe coents on 8 and 9 are in order. Condition 8 is necessary for 6 to be well-posed. Indeed, consider any 0, w 0 K,M such that 0 has copact support. An easy coputation shows that if α > /, then Eσ 0 σ, σ + w 0 σ as σ 0, so E is not bounded fro below on K,M. We show that 8 is indeed sufficient for e M to be finite, and allows to look for ground states of. This will be accoplished by a study of the Sobolev regularity of the Kologorov equation, see in particular Section 2.2. ote that the critical case α = / is ore delicate, and require additional analysis. We also ention that another critical exponent is intrinsic in : if α > /, one has to expect non-existence of solutions see [3]. 3

4 As for condition 9, it guarantees Hölder regularity of iniizers. Basically, the integrability of a iniizing sequence n, w n of E increases as increases. By Corollary 2.9, iniizers are Hölder continuous if 9 is in force. In the case <, one has to expect just L p a-priori regularity of, and soe additional work is required to conclude the existence of a classical solution to. This study is atter of a work in progress. The ai of this work is two-fold. Firstly, for any fixed > 0, we prove the existence of classical ground states of. Secondly, we study the behavior of solutions in the vanishing viscosity liit 0. Copactness of iniizing sequences of E is guaranteed by the coercivity of V and Hölder regularity. Then, the existence of a solution u, λ of the HJB equation in is obtained by considering another functional with linearized coupling around the iniizer and the associated dual functional in the sense of Fenchel-Rockafellar as in [9]. One has to take care of the interplay between u and as x. The behavior of value function u is controlled by Bernstein estiates, while a control on the decay of coes fro the assuptions on V. ote that we will consider classical solutions to this syste with a slight abuse of terinology, that is u, C 2,β W,p, for soe β 0, and all p >. The existence result is stated as follows. Theore.. Suppose that 2, 7, 8 and 9 hold. Then, for every > 0 and M > 0, i There exists a iniizer, w K,M of E, that is E, w = inf E, w.,w K,M ii For any iniizer, w K,M classical solution to. of E, there exists u, λ such that u, λ, is a In the second part of the work, we analyze the behavior of the triple u, λ, coing fro a iniizer of E as 0. Fro the viewpoint of the odel, this aounts to reove the Brownian noise fro the agents dynaics. Heuristically, if the diffusion becoes negligible, one should observe aggregation of players induced by the decreasing onotonicity of coupling in the cost towards inia of the potential V, that are the preferred sites. Fro the PDE viewpoint, this phenoenon should appear in the concentration of the density around a finite nuber of points. In order to bring as uch as possible inforation to the liit, we consider an appropriate rescaling of, u, naely = α α +x, ũ = αγ α u α +x ux, 0 for all > 0. The rescaling is designed so that ũ, solves a MFG syste where the nonlinearities have the sae behavior of the original ones, i.e. H p γ as p, but the coefficient in front of the Laplacian is equal to one for all, see 73. Moreover, the couple ũ, is associated to a iniizer of a rescaled energy E, see 69. It turns out that in this rescaling process, the potential V becoes V = α α V α, and vanishes locally as 0. Therefore, as one passes to the liit, the potential cannot copensate anyore the lack of copactness of, and the convergence of in L has to be proven by other ethods. Heuristically, the aggregating force should be strong enough to overcoe the dissipation effect, but the clustering point can be hard to predict by lack of spatial preference. This is why we also have to translate in 0 by x. We will select x to be the iniu of u : heuristically, being u the value function, this is the point where ost of the players should be located. In order to recover copactness for the sequence, we ipleent soe ideas of the celebrated concentration-copactness ethod [27]. This principle states intuitively that if loss of 4

5 copactness occurs, splits in at least two parts which are going infinitely far away fro each other, that is χ BR 0 + χ \B 2R 0, with R, χ BR 0 a and χ \B 2R 0 M a for soe a 0, M a third possibility ight happen, but it is easily ruled out here by local estiates. This induces a splitting in the energy E, that is inf E inf E + inf E. 2 =M =a =M a One then exploits a special feature of E, which is called sub-additivity: inf E < =M inf E + inf E, =a =M a that akes 2 ipossible. While sub-additivity is easy to prove for E see Lea 4.5, the splitting 2 requires technical work, in particular due to the presence of the ter L w/ in E, that becoes increasingly singular as approaches zero a siple cut-off as in is not useful. The property 2 is proven in Theore 4.6. It relies on the Brezis-Lieb lea and a perturbation arguent. The L convergence of enables us to obtain the full convergence of ũ, to a liit MFG syste. By a unifor control of the decay of as x, that coes fro a Lyapunov function built upon ũ, and energy arguents, we are also able to keep track of x. In ters of the original density, x is the point around which ost of the ass is located. For technical reasons, we require additional hypotheses on V and H. We will assue that the potential V has a finite nuber of inia and polynoial behavior, that is for soe ˆb > 0, x j, j =,..., n, V x = hx n j= x x j ˆb, C V hx C V on. 3 note that nˆb = b, in assuption 2. Moreover, we will strengthen the assuptions on the Hailtonian of the syste: in particular we suppose that H C 0,γ loc 4 and we will need that either H is hoogeneous or the potential has a growth at infinitity controlled by γ: either Hp = C H p γ or b = nˆb < γ 2. 5 Actually, ost of the analysis can be carried over under the ore general assuptions 7, see the discussion at the beginning of Section 4. The second ain result of this work is stated as follows: Theore.2. Assue that 3, 7, 4 and 5 hold. Let u, λ, be as in Theore.. Then, there exist sequences 0 and x, such that for all η > 0 there exists R and 0 for which for all < 0, dx M η, 6 and for soe J =,..., n, C > 0, x x R α x x J C n α. 7 Moreover, α γ α u γ α +x, αγ α γ α +x converges respectively in Cloc R and in C loc L p for all p to a classical solution of u + H 0 u + λ = α div H 0 u = 0 8 = M, 5

6 where λ = li 0 α α λ and H 0 p = li 0 H p. Theore.2 describes the liiting behaviour of u, : in particular, concentrates around inia of V. Moreover, it states the existence of a solution to an ergodic MFG syste with decreasing coupling α, without coercive potential. In ters of the MFG odel, a stable longtie equilibriu is possible by the intrinsic aggregating force even without a spatial preference of the players. ote that in the very special case of quadratic Hailtonian, that is Hp = p 2 /2, the change of variables vx = C exp ux/2, x = v 2 x, transfors the syste into { 2 v + V x λv = v 2α+ v 2 = M. Concentration phenoena of positive ground states as 0 of the previous seilinear equation have been widely studied, in different settings. We just recall [8] where a siilar proble with ass constraint in diension = 2 and with α = is considered, see also references therein. We ake a final reark regarding soe results in [5], where a first-order one-diensional proble with aggregation is considered. In their odels, it happens that the focusing effect ay prevail the spatial preferences of the agents, naely the density has support far fro inia of V. It sees that such explicit solutions are not selected by liits of our viscous odels, that can be only singular easures. The paper is organized as follows. In Section 2 we provide soe regularity results on the Hailton-Jacobi-Bellan and Kologorov equation that will be used thoroughly. Mainly, they coe fro results in [5, 0, 3]. Section 3 is devoted to the proof of a solution to the MFG syste when > 0 is fixed. In Section 4, we study the vanishing viscosity liit 0: we prove copactness of the rescaled density and prove the full convergence of, ũ to ground states of a MFG syste which is potential-free. Acknowledgeents. The authors are partially supported by the Fondazione CaRiPaRo Project onlinear Partial Differential Equations: Asyptotic Probles and Mean-Field Gaes and PRAT CPDA57835 of University of Padova Mean-Field Gaes and onlinear PDEs. A.C. is partially supported by the IdAM-GAMPA project Tecniche EDP, dinaiche e probabilistiche per lo studio di problei asintotici. 2 Soe preliinary regularity results otations. For every p, p = p p will be the usual conjugate exponent of p. For all R > 0, x, B R x := {y : x y < R}. We will denote by ω := B 0. Finally, C, C, K, K,... denote positive constants we need not to specify. The assuptions on H guarantee the following see, e.g., [2, Proposition 2.] Proposition 2.. There exist C L, C, C 2 > 0 depending on C H and on γ such that p, q, i L C 2 \ {0} and it is strictly convex, ii 0 C L q γ Lq C L q γ +, iii Lq q Lq C q γ C, iv C q γ C Lq C q γ +. v C 2 p γ C 2 Hp C 2 p γ +. We will use the following standard result on Hölder functions vanishing at infinity. 6

7 Lea 2.2. Suppose that 0, C 0,θ c h, for soe θ, c h > 0, and dx <. Then, x 0 as x. Moreover, if dx < η x R for soe η, R > 0, then where C > 0 depends only on c h,. θ ax x Cη θ+, 9 x R Proof. By contradiction, suppose that there exists δ > 0 and a sequence x n such that x n > δ for all n. We ay also assue that x n+ x n + for all n. By the Hölder regularity assuption, provided that x B r x n, and r θ for all n. Then, dx n x x n c h x x n θ δ 2, δ 2c h. Choose r = in{, B rx n dx n θ δ 2c h δ 2 B r0 = + }, so that B r x n B r x = that is ipossible. As for the second part, let M := ax x R x = x, x R note that such a axiu is achieved as a consequence of the first part of the lea. As before, x x c h x x θ M 2 /θ. M for all x B r x, where r = 2c h Therefore, and 9 follows. η > x R dx M 4 B r x = M /θ M 4 B 0, 2c h We recall the following well known result, proved in [8, Theore ]. Theore 2.3. Let f n f a.e. in and assue that f n L p C for all n and for soe p [, +. Then li n [ f n p L p f n f p L p ] = f p L p. Fro classical elliptic regularity, we have the following result. Proposition 2.4. Let p > and L p be such that ϕ dx K ϕ L p for all ϕ C0 for soe K > 0. Then, W,p and there exists C > 0 depending only on p, such that L p C K. 7

8 Proof. Fix any R >. Let ψ C 0 B 2 0, ϕrx := ψx so, ϕ C 0 B 2R 0 and vx := Rx on. Then, B 20 v ψ dx = R2 B 2R 0 ϕ dy KR2 ϕ p dy B 2R 0 = KR +/p ψ p dx B 20 /p /p KR /p ψ W,p B 20. Hence, by [, Theore 6.], v W,p B 0 and there exists a constant C, depending on p but not on R, such that Therefore, v L p B 0 v W,p B 0 CKR /p + v L p B 20. /p p dy = /p B R 0 B 0 v p dx /p C Letting R, we get that L p R n and the desired estiate. K + /p v p dx B 20 /p = CK + R L p B 2R The Hailton-Jacobi-Bellan equation on the whole space In this section we provide soe a-priori regularity estiates and existence results for Hailton- Jacobi-Bellan equations in the whole spaces of ergodic type. In particular we will consider failies of Hailton-Jacobi-Bellan equations u n + H n u n + λ n = F n x f n x on 20 where f n C 0 L, λ n R are equibounded in n, that is λ n λ and f n C f with C f independent of n. Moreover H n is for every n an Hailtonian which satisfies 7, with constants γ and C H independent of n; finally, there exists C F 0 and b 0 independent of n such that C F ax{ x C F, 0} b F n x C F + x b n and x. 2 ote that, differently fro assuption 2 for the potential V, the function F n can also be bounded, if b = 0. Theore 2.5. Let u n C be a sequence of viscosity solutions of the HJB equations 20. Then there exists a constant K > 0 depending on C H, C F, C f, γ,, λ such that u n x K + x b γ, 22 where b 0 is the growth of F n appearing in 2 and γ is the growth of H n appearing in 7. Proof. Without loss of generality we ay consider H n p = C H p γ for all n and p. Indeed, every v n solves u n + C H u n γ + λ n = F n x f n x + C H u n γ H n u n on, and since C H u n γ H n u n C H by 7, we can redefine f n to include C H u n γ H n u n, which then satisfies the bound f n C f + C H. ote that it is also sufficient to prove the theore for u n C 2, as the general case follows by approxiation. 8

9 We first clai that if v C 2 B 2 0 satisfies for soe k > 0, then we have for any r [, ], v + C H v γ k on B 2 0 v Lr B 0 C, 23 where C depends only on k, C H, γ,, r. If r [,, this is proven in [22, Theore A.]. The case r = follows by classical elliptic regularity, since if r in 23 is large enough, then v is bounded in L q B 3/2 0 for soe q >, and the stateent follows by Sobolev ebeddings. In view of these considerations, the gradient bound 22 easily follows if b = 0. For the case b > 0, fix x 0, and let δ = + x 0 b/γ. Let Then, v n solves Since δ, v n y := δ 2 γ γ un x 0 + δy on. v n + C H v n γ = δ γ F n x 0 + δy f n x 0 + δy λ n. δ γ F n x 0 + δy f n x 0 + δy λ n C F 3 + x 0 b + C f + λ + x 0 b C for all y B 2 0 by 2 and the bound on f n. Therefore, by the first clai, v n L B 0 C, for all n. In particular, choosing y = 0, and the desired estiate follows. u n x 0 = δ γ vn 0 C + x 0 b/γ, Moreover, we prove the following a-priori estiates on bounded fro below solutions to 20. Theore 2.6. Let u n C be a faily of uniforly bounded fro below viscosity solution to 20, that is for which there exists C > 0 such that u n C for every n. If b = 0 in 2, we oreover assue that there exists η > 0 and R > 0 independent of n such that F n x f n x λ n > η > 0, for all x > R. 24 Then there exists C > 0 such that u n x C x + b γ C, n, x, 25 where b 0 is the growth power appearing in 2 and γ is the growth power appearing in 7. Proof. The proof is based on the sae arguent as in [5, Proposition 3.4], we sketch it briefly for copleteness. Since u n is bounded fro below we can assue u n 0, up to adddition of constants. We assue by contradiction that 25 does not hold. Then there exist sequences x l and u nl, such that x l > 2R, x l +, and un l x l 0. Let a l = x l 2 and we define the function x l + b γ v l x = a + b γ l 9 u nl x l + a l x.

10 By Theore 2.5, we get that u nl x K+ x b γ. Therefore, v l, v l are uniforly bounded. Moreover, v l is a solution to a b γ l v l + H nl a b γ l vl + λ nl = F nl x l + a l x f nl x l + a l x. In particular, recalling 7, we get that v l is a supersolution to a b γ b l v l + C H v l γ a b l λ nl + F nl x l + a l x f nl x l + a l x. ote that, for every l sufficiently large, by 2 and by 24 in the case b = 0 the right hand side of the equation a b l λ nl + F nl x l + a l x f nl x l + a l x > 0 for x such that x. Moreover, passing eventually to a subsequence, we get that v l v locally uniforly in n and 0. So v is a supersolution to C H v γ η > 0 in B0, with hoogeneous boundary conditions since v 0. By coparison, recalling the explicit forula of the solution to the eikonal equation f γ = C in B0, with hoogeneous boundary conditions, we conclude that vx C γ x for all x such that x. Moreover, by unifor convergence, we get that, eventually enlarging C and taking l sufficiently large, v l x C γ x for all x with x, in particular v l 0 C γ. Recalling the definition of v l, we get that v l 0 0, which yields a contradiction. a b γ b l Define λ n := sup{λ R : 20 has a solution u n C 2 }. Theore 2.7. Assue that for every n the function F n f n is locally Hölder continuous and bounded fro below uniforly in n. i λ n <, for every n, and there exists, for every n, a solution u n C 2 to 20 with λ n = λ n. Moreover λ n := sup{λ R : 20 has a subsolution u n C 2 }. ii If F n satisfies 2, with b > 0, then, for every n, the solution u n to 20 with λ n = λ n is unique up to addition of constants and satisfies 25. iii If F n 0, and there exists δ > 0 independent of n such that li sup f n x + λ n < δ < 0, 26 x + then for every n there exists a solution to 20 with λ n = λ n which satisfies 25 with b = 0. Proof. i. The proof of this result can be obtained by a straightforward adaptation of the proof of Theore 2. in [5], using the a-priori estiates on the gradient given in Theore 2.5. Observe that actually in [5] it is required a stronger assuption on F n f n, that is it is required it is locally Lipschitz continuous. This assuption is used to derive a-priori estiates on the gradient of solutions by using the so called Bernstein ethod see Appendix A in [5], which depends also on the L nor of F n f n. In our case we can weaken this assuption to just Hölder continuity so still ensuring classical elliptic regularity since we are using a-priori estiates on the gradient given in Theore 2.5, which depends only on the L nor of F n f n, and are obtained in [22] by the so called integral Bernstein ethod. ii. For the proof we refer to [20] see also [5] and [2]. In particular in [20], it is proved that u n is bounded fro below. By looking at the proof, it is easy to check that, due to the unifority in n of the nors of coefficients, the bound can be taken independent of n, then by Theore 2.6, we get the estiate on the growth. 0

11 iii. By adapting the arguent in [5, Theore 2.6], we get that there exists a bounded fro below solution to 20 with λ n = λ n, with bound unifor in n. Then using Theore 2.6, we get the estiate on the growth. We give a brief sketch of the proof of the existence of a bounded fro below solution. For every R > 0, we consider the ergodic proble { u R n + H n u R n + λ R n = f x < R u R 27 n x + x R. Using the result in [4], we get that for every R > 0 there exists a unique λ R n and a unique up to addition of constant solution u R n C 2 B R. First of all we clai that li R λ R n = λ n. It is easy to check that if R > R, then λ R n λ R n, and oreover that λ R n λ n. So, the sequence λ R n is converging as R + to soe λ n λ n. Moreover, by the sae arguent as in Theore 2.5, we get that for every copact K, there exists a constant C > 0 such that u R n C in K for every R sufficiently large and for all n. Without loss of generality we can assue that u R n 0 = 0 for every R. So, using the gradient bound, and elliptic regularity, we conclude that u R n is bounded in C 2 K by soe constant independent of R. Hence, by Ascoli-Arzelà Theore, and via a diagonalization procedure, we get that u R n converges locally in, with u n C 2. Moreover, u n is a solution to 20, with λ = λ n. Recalling the characterization of λ n and the fact that λ n λ n, we conclude that λ n = λ n. Then, we consider x R n B R such that u R n x R n = in x R u R n. Recalling that u R n is a solution to 27, we get by coputing the equation at x R n and by recalling that H n 0 0, that λ R n + fx R n H n 0 + λ R n + fx R n 0. Using condition 26, and recalling that λ R n λ n, we get that there exists a copact set K independent of R and of n and R 0 > 0 such that for all R > R 0, x R n K. Recalling that u R n 0 = 0 and u R n C in K with C independent of n, R, we conclude that u R n x R C for soe constant C independent of n, R. But, this iplies, since u R n x u R n x R n for every R, that passing to the liit u n x C, with C independent of n. 2.2 A-priori estiates for the Kologorov equation In this section we do not assue > and we provide general a-priori estiates for couples, w L W,q L such that x = M and + div w = 0 where q = { + <. Lea 2.8. Let β q q, for q <, and β < + for q. We define r β as follows r = + β. 29 Then, there exists a constant C, depending only on and β, such that W,r C γ CL C γ w L γ 28 dx + M γ L β 30 w dx + M γ L β, where C L is the constant appearing in Proposition 2.. We now assue that < β < + γ. 3

12 Then, there exists δ > 0 such that +δβ L β C M +δβ γ w γ dx where the constant C depends only on γ,, and β. CC L M +δβ γ L w dx, 32 Proof. Since W,q, by Sobolev ebedding and interpolation, we get that L β. Using + div w = 0, we get for all ϕ C0, ϕ dx = w ϕ dx. Using Holder inequality, recalling 29, we obtain w ϕ dx w ϕ dx R Therefore, we get that for all ϕ C 0, ϕ dx γ γ w γ dx w γ dx γ L β ϕ L r. γ L β ϕ r. We apply then Proposition 2.4 and we obtain that W,r and that there exists a constant C, depending only on r, such that L r C γ w γ dx γ L β. 33 Fro this inequality, using Proposition 2. and recalling that by interpolation, since L = M, L r γ L β M, we conclude the desired inequality 30. ow we fix η such that η = r = ote that, by a siple coputation using 29, we get η β = by 3, we conclude that that η > β. = M, we get + r. + β β γ, therefore, By Gagliardo irenberg inequality, and recalling that L η C + L r M Since η > β, by interpolation we get that there exists θ > such that θ L β L η M θ. Actually θ = +. β + β So, we substitute in 34 and 33 and we get, elevating both ters to +, + θγ L β C M θ + R w γ γ dx γ L β. 35 ow, since θ >, by 3, we get θ + γ γ = β β = β + β [ ] γ β + β > 0. 2

13 Therefore we deduce 32 fro 35 with δ = β [ γ + β ]. 36 Corollary 2.9. For every r < q, there exists C > 0 depending on, and r such that W,r C C L L R w dx + γ M. 37 γ Moreover, if > so q >, then C 0,θ and C 0,θ C C L L R w dx + γ M. 38 γ Proof. For q equivalently, we fix r < q and we choose β which satisfies 29 for such r. By Sobolev ebedding theore, W,r is continuously ebedded in L β. So, there exists C depending on and r such that L β C W,r. Using inequality 30, we get L β C R w γ dx + γ M. γ If we substitute again in 30 we get W,r C R w γ γ dx + γ M. In particular for q >, we can choose r > and by Sobolev ebedding theore we get that there exists θ = r and a constant C > 0 depending on and r such that C 0,θ C R w γ dx + γ M γ C C L L R w dx + γ M. γ For q <, we fix r < q, and choose the corresponding β in 29, that satisfies β < Hence we conclude again fro inequality 30.. Corollary 2.0. Assue that u, W,γ W,γ L solves the Kologorov equation in. For any β satisfying 3, there exists δ > 0 such that +δβ L β C u γ dx γ where C is a constant which depends only on, β, γ. Proof. We apply Lea 2.8 with w = H u. ote that w H u C H u γ +. So, since u L γ and L γ we obtain w L γ. 3

14 3 Existence of solutions to the MFG syste for > 0 3. A-priori estiates and energy bounds In this section, we provide bounds fro above and below for the energy E, assuring in particular that the iniu proble is well defined. Lea 3.. Let, w K,M. Then E, w K C α α 39 where C, K > 0 are constants depending only on, M, C L, γ, α, M, V. In particular there exists finite e M = inf E, w.,w K,M Proof. Recalling that V 0 and applying the estiate 32 with α = β, we get E, w L w dx α+ dx R α + C γ M +δ+α +α+δ L α+ α + +α L α+ + Cδ γ δ δ δ + α + where C is a constant depending only on, M, C L, γ, α and δ = [ ] γ α α. 40 Thefore, substituting in the energy, we get E, w C γ α α which gives the desired inequality. Lea 3.2. There holds α α α α, α + α inf E, w C 2 α K2 4,w K,M where C 2 > 0, K 2 are constants depending only on, M, C L, γ, α, V. Proof. We construct a couple, w K,M as follows. First of all we consider a sooth function φ : [0, + R which solves the following ordinary differential equation { φ r = φr + φr α φ0 = 2. Then, it is easy to check that 0 < φr 2 e r. We define x = Aφτ x, where A, τ are constants to be fixed, and wx = x. First of all we ipose M = xdx = A R τ φ y dy = A R τ C, 4

15 recalling that φ is exponentially decreasing. So A = Mτ C, where C = φ y dy. Observe also that α+ xdx = M α+ τ α C α+ φ α+ y dy = M α+ τ α C α+ C α 42 where C α = φ α+ y dy. We check, recalling that the growth condition 2, that the following holds y xv xdx = MC V φ y dy = C R τ τ b, 43 where K is a constant depending on, φ, C 0. Moreover, we copute, recalling that φ solves the ODE w γ = τ + γ M α C α α = γ τ γ γ + τ α M α C α α. 44 τ α We consider the energy at, w E, w = L w dx α+ dx + V xdx. R α + Using Proposition 2., and coputation 44 and 42, we get L w dx α+ dx R α + R C L w γ dx + C L M α+ dx R γ α + R = C L γ τ M γ + R M α C α τ α α+ dx + C L M α + τ = C L γ τ γ M + C L M α + γ γ α M α C α α+ dx = MC L + MCC α γ τ γ α + M α+ C α+ C α τ α + C L M. α+ We choose now τ such that τ = K α, where K is sufficiently large, in such a way that L w dx α+ dx C α α + CL M R α + where C is a constant depending on α, C L, M. Substituting this in the energy and recalling 43, we get the desired inequality. We get also a-priori bounds on iniizers and iniizing sequences. Proposition 3.3. Let, w K,M such that e M E, w η, for soe positive η. Then w γ dx C α α + K, 45 L α+ α C α + K, 46 α+ C 0,θ C 2 α + K, 47 for soe θ 0,, C, K positive constants which depends only on α,, V, C L. 5

16 Proof. ote that if, w K,M, then, θw K θ,m. Let, w K,M such that e M E, w + η, for soe positive η. Then we get, [ e M e θ M E, w + η E, θw = L w L θ w ] dx + η. 48 Recalling the properties of L in Proposition 2. we get [ L w L θ w ] dx MC L + C L θ γ We choose 0 < θ < so that θ γ > 0. Using 39 and 4, we get e M e θ M K + C θ α α γ α α = K K 2 + α α w γ dx. 49 K2 C 2 γ α α 50 α α C2 C θ γ = K + C α α, where, since C > C 2 and θ <, we get that C = C θ α α C2 > 0. Substituting 50 and 49 in 48, we conclude 45. Fro 45, using the estiate 32, with β = α + and δ + = γ α using 36, we get α+ L α+ C α w γ dx α C α C γ as it can be coputed α α + K α fro which we conclude 46. Finally, by 38 proved in Corollary 2.9, we deduce the a-priori bound on C 0,θ. 3.2 Existence of iniizers of E and solutions to for every > 0. We are now in the position to show existence of iniizers of the energy E in the class K,M for every, M > 0. Theore 3.4. For every > 0 and M > 0, there exists a iniizer, w K,M of E, that is E, w = inf E, w.,w K,M Moreover, for every iniizer, w K,M of E, there holds + x b L, w + x b/γ L, w L q for all q, 5 and there exist constants C > 0 and K, independent of, such that w dx + α+ L α+ α C α + K. 52 Proof. Let n, w n K,M be a iniizing sequence, that is E n, w n e M. This iplies that, choosing n sufficiently large, E n, w n e M +. Fro this we get n L w n dx + V x n dx E n, w n + n R α + α+ n dx e M + + α + α+ n. 53 6

17 Proceeding as in Proposition 3.3, we get that both w n γ dx and n C 0,θ C are γ n bounded by soe constant C depending on, e M, M,,. Eventually reducing θ and using Ascoli-Arzelá theore we get that, up to subsequences, n in C 0,θ. By interpolation between L and L, we obtain that n is also bounded uniforly in L α+, and hence, using again 53 and the assuptions on V, we have + x b L. Observe oreover that fro 53, we get that V x n dx C, where C is independent of n. Using the positivity of V and n and the assuption 2, we deduce fro that, for every R > C V, x R n xdx C in x R V x C C V R C V b. This iplies, by unifor convergence, that for every η there exists R such that xdx M η. B R 0 Using a diagonal arguent, we can prove that, eventually passing a subsequence, n in L and xdx = M. A consequence of the convergence of n is that there exists C > 0 such that n C on for all n. Therefore, C L w C γ n dx n L w n dx, R n that is bounded uniforly in n by 53 and Proposition 3.3. Thus, up to subsequences, w n w in L γ. ote that w dx w + x b/γ dx w γ /γ /γ dx + x b dx, and so we deduce that w L q for all q by interpolation and that w + x b/γ L. The convergence is strong enough to guarantee that, w K,M. Since E is lower seicontinuous on K,M,, w is a iniizer of E. Theore 3.5. Let, w K,M be a iniizer of E. Then, there exists u, λ such that u, λ, is a classical solution to u + H u + λ = α + V x div H u = 0 54 = M, and w = H u. Moreover, there exist constants K, K 2, C, C 2, C > 0, independent of, such that K C α α λ K 2 C 2 γ α α 55 u x C x b/γ+ C, u x C + x b/γ on. 56 7

18 Proof. Let, w be a iniizer of E. Define the space of test functions { A = A b,γ := ψ C 2 ψx ψx : li sup <, li sup x x b/γ x x b ote that we also have, for all ψ A, We clai that li sup x ψx <. x b/γ+ < }. 57 ψ dx = w ψ dx ψ A. 58 Indeed, consider a radial sooth cutoff function χx which is identically equal to one in B 0 and identically zero in \B 2 0. Set χ R x := χx/r; we have χ R C R and χ R C R 2 on for soe positive constant C. Since the equality = divw holds in the weak sense on, we ay ultiply it by χ R ψ with ψ A and integrate by parts to obtain χ R ψ + 2 ψ χ R + ψ χ R dx = w χ R ψ + ψ χ R. dx 59 B 2R B 2R ote that for soe positive C, w ψ dx C w + x b/γ dx <, by the integrability properties 5. Moreover, R x 2R ψ dx C + x b dx < + x b/γ+ ψ χ R dx C R x 2R R 2 dx C + x b/γ dx 0 as R, R x 2R because b/γ b. Reasoning in a siilar way, we also have that R x 2R ψ χ R and R x 2R w ψ χ R converge to zero as R. Equality 58 then follows by passing to the liit in 59. Therefore, the proble of iniizing E on K,M is equivalent to iniize E on K, where K := {w, L W,γ L γ : w, satisfies 5, 58, 0 and = M} As in [9, Proposition 3.], convexity of L iplies that, w is also a iniizer of the following convex functional on K: J, w = Ψ, w + V x α dx. We now ai to prove that sup{λm : ψ + H ψ + λ V x α on for soe ψ A} = in J, w. 60 w, K We proceed as in [, Theore 3.5]: setting L, w, λ, ψ := J, w + ψ + w ψ λ dx + λm, 8

19 we have in J, w = in,w K,w sup λ,ψ R A L, w, λ, ψ, where the iniu in the right hand side has to be intended aong couples, w L W,γ L γ satisfying 5. ote that L,, λ, ψ is convex, and L, w,, is linear. Moreover, since L,, λ, ψ is weak-* lower sei-continuous, we can use the in-ax theore see [7, Theore 2.3.7], to get in sup,w λ,ψ R A sup λ,ψ R A,w sup λ,ψ R A L, w, λ, ψ = in L R w in L,w R sup λ,ψ R A,w in L, w, λ, ψ = + V x α + ψ + w ψ λ dx + λm = w + V x α + ψ + w ψ λ dx + λm, where the interchange of the in and the integration is possible by standard results in convex optiisation. By coputation, in,w R L w + V x α + ψ + w ψ λ is zero whenever ψ H ψ λ + V x α is positive, and it is otherwise. Therefore, we have proven 60. By Theore 2.7, i, ii, there exists u C 2 such that u + H u + λ = V x α on, 6 and which satisfies the first estiate in 56. Moreover, by the gradient estiates in Theore 2.5 with F = V and f = α we have that and u x C + x b/γ+, u x C + x b/γ on u x H u x + λ + V x + α x C + x b on by 7, so u A. Thus, the supreu in the left hand side of 60 is achieved by λ, and it holds true that λ M = J, w. 62 This gives in particular 55, using Leata 3., 3.2, and 52, recalling Proposition 3.3. We now use 62, 6 and 58 with ψ = u to get 0 = that iplies L w + V x α λ dx = = L L w w w = H u on the set { > 0}. u + H u dx + H u + u w dx, Hence, the Kologorov equation + div H u = 0 holds in the weak sense, and by elliptic regularity we conclude that u, λ, is a classical solution to. We conclude with the proof of the first ain result of this paper. Proof of Theore.. Follows by Theores 3.4 and

20 4 Convergence to ground states and concentration In the second part of the work we analyse the behaviour of the syste as 0. Our ai is to show that concentration phenoena happen, and also to describe the asyptotic shape of the liits in ters of ground states of the syste. As stated in the introduction, we will suppose that fro now on 3, 7 and 5 hold. We ention that Proposition 4.3 and Theore 4.6 could be proved under the ore general assuptions of Theore.. On the other hand, if only 7 is satisfied, soe ore work is required to prove the estiates in Proposition 4.0 and then to conclude the full convergence result. So, we decided to pass to the stronger assuptions 7, 5 in order not to increase too uch the technicalities in the paper. 4. The rescaled proble Given x to be chosen later see 72 below, we consider the following rescaling y := α α y + x, wy := +α α w α y + x ũy := α γ α u α y + x ux λ := αγ α λ. 63 ote that, w K,M if and only if, w K,M =: K M, that is 0, dx = M and ϕ dx = w ϕ dx ϕ C0. We introduce the rescaled potential note that we do not translate by x here V y = α α h and the rescaled Lagrangian and Hailtonian L q := α α L By 7, there exists C > 0 such that So, we get that α α q, α y n j= α y xj ˆb, 64 H p = αγ α H α α p C L q γ L q C L q γ + C L α α, 66 C q γ C α α L q C q γ + C αγ α, C H p γ α α CH H p C H p γ, 67 C p γ α α C H p C p γ + C α α. li L q = L 0 q := C L q γ, 0 li H p = H 0 p := C H p γ uniforly in Moreover, 4 iplies that H is locally bounded in C 0,γ, and therefore that also H p H 0 p = C H γ p γ 2 p locally uniforly. We rescale the energy using 63: E, w = α α L w + V y + y α + α+ dy =: α α E, w, 69 20

21 where and the iniization proble 6 is therefore equivalent to ẽ M = y = α x, 70 inf E, w., w K M Recalling Leata 3., 3.2, we get that there exists C, C 2 > 0 and K, K 2, independent of, such that C K α α inf E, w C 2 K 2 α α. 7, w K M Let, w be a iniizer of E and u, λ, be the associated solution to 54, as defined in Theore 3.5. Then by 56, there exists x such that u x = in u x. 72 We fix such x in 63 and we define the corresponding y according to 70. An iediate corollary of Theore 3.4 is the following. Corollary 4.. For every > 0 there exist a iniizer, w K M to E. Moreover, there exists ũ, λ such that ũ 0 = ũ 0 and ũ, λ, is a classical solution to ũ + H ũ + λ = α + V y + y div H ũ = 0 73 = M, and w = H ũ. Finally there exist θ 0,, C, C, C 2 > 0 independent of such that C λ C 2, γ w dx + α+ L α+ C, C 0,θ C. 74 Proof. The corollary directly follows by Theore 3.4 and Theore 3.5, taking into account the previous rescalings 63, with the choice of x defined in 72. Estiate 74 is a consequence of 55 and 52. Finally the Hölder estiate is obtained by applying the sae arguents as in Corollary 2.9 and Lea A preliinary convergence result In this section, we provide soe preliinary convergence results, where we are not preventing possible loss of ass in the liit. First of all we need soe a-priori estiates on the solution to 73. Lea 4.2. Let, w K M be a iniizer to E as in Corollary 4., and let ũ, λ, be the associated classical solution to 73. Then, there exist constants K, C, K > 0, independent of, such that xdx K, R > and ũ y C + y nˆb γ, 75 B R 0 α+nˆb α y nˆb C and V y + y K α+nˆb α y nˆb

22 Proof. By the rescaling 63 and the choice of x in 72, we have that 0 = ũ 0 = in ũ. Since ũ is a classical solution to 73, we get, coputing the equation in 0, H 0 + λ α 0 + V y. Recalling that H 0 0 and V 0, we get that α 0 λ C 2 > 0, where C 2 is the constant appearing in 74. Since is uniforly bounded in C 0,θ, as stated in 74, and 0, there exists a constant K depending on θ, C, C 2 as in 74, such that for all R > we have xdx K > B R 0 Up to passing to a subsequence we get that locally uniforly, where 0 is in C 0,θ, with θ < θ and with Hölder nor bounded. Moreover, due to 77, by Fatou lea and unifor convergence in copact sets, we obtain M dx K > 0. Using the fact that is a iniizer of E, we obtain that there exists a constant, independent of, such that V x + y xdx C. This iplies, using 64 and 77, that for all R >, we have, letting K x i for every i, K α α C B R 0 V x + y xdx K α α n α y γ α R xj j= ˆb inf n x B R 0 j= K α α α x + α y x j ˆb α y γ α R K nˆb K α+nˆb α y nˆb K αγ α. 78 Therefore, 78 iplies that there exists a constant C > 0 such that α+nˆb α y nˆb C. Recalling 64 we get that there exist constants K 0, K, independent of, such that 0 V y + x K 0 α+nˆb α y nˆb + K 0 α+nˆb α x nˆb + K 0 αγ α K α+nˆb α x nˆb +, which gives 76. Finally, the gradient estiate in 75 coes fro 76 and Theore 2.5. Proposition 4.3. Let, w K M be a iniizer to E as in Corollary 4., and let ũ, λ, be the associated classical solution to 73. Up to subsequences, we get that λ λ, and ũ ū,, ũ ū, H ũ H 0 ū 79 locally uniforly, where ū 0 = ū0, and ū, λ, is a classical solution to { ū + H 0 ū + λ = α + gx div H 0 ū = 0 80 for a continuous function g such that 0 gx C on for soe C > 0. Moreover, there exists a 0, M] such that R n dx = a, 8 22

23 and C, K, κ > 0 such that ūx C x C, ū K on, e κ x xdx < Proof. First of all observe that, since V is a locally Hölder continuous function, then 76 iplies that, up to subsequence, V x + y gx, locally uniforly as 0, where g is a continuous function such that 0 gx C, for soe C > 0. Using the a-priori estiate 75, 74 and 76, and recalling that ũ is a classical solution to 73, by classical elliptic regularity theory we obtain that ũ is locally bounded in C,α in every copact set, uniforly with respect to. So, up to extracting a subsequence via a diagonalization procedure, we get that ũ ū, ũ ū, locally uniforly, and λ λ. Moreover, also H ũ H 0 ū locally uniforly. So, we can pass to the liit in 73 and obtain that ū, λ, ū is a solution to 80, which is classical by elliptic regularity theory. Observe that ū is a solution to ū + H 0 ū + λ = α + gx. By Theore 2.5, we get that there exists a constant K depending on sup g and λ such that ū K. Moreover, by construction ū 0. Since is Hölder continuous, and such that dx = a 0, M], by Lea 2.2, we get that 0 as x +. Therefore, we get that li inf x + α x + gx λ H 0 0 λ > 0. So, by Theore 2.6, recalling that by construction ū0 = 0 ūy, we get that ū satisfies ūx C x C 83 for soe C > 0. To conclude, consider the function Φx = e κūx. We clai that we can choose κ > 0 such that there exist R > 0 and δ > 0 with Φ + H 0 ū Φ > δφ x > R. 84 Indeed, since ū solves the first equation in 80, we get Φ + H 0 ū Φ κ λ κ ū 2 α Φ. Using 83 and 0 as x +, we obtain the clai. Reasoning as in Proposition 4.3 in [2], we get that e κū dx < +, which concludes the estiate 82. Reark 4.4. With estiates 82 in force, the pointwise bounds stated in [29, Theore 6.] hold, naely there exist positive constants c, c 2, such that 4.3 Concentration-copactness x c e c2 x on. In this section we show that actually there is no loss of ass when passing to the liit as in Proposition 4.3. In order to do so, we apply a kind of concentration-copactness arguent. First of all we show that the functional E, w enjoys the following subadditivity property. Let us denote ẽ M = in,w KM E, w. 23

24 Lea 4.5. For all a 0, M, there exists a constant C = Ca, M 0 depending only on a, M and the data not on, such that CM, M = 0 = C0, M, Ca, M > 0 for 0 < a < M and ẽ M ẽ a + ẽ M a Ca, M. 85 Proof. Let c > 0. For all adissible couples, w K M we have E c, cw = cl w cα+ α + α+ + cv x + y dx If, in addition, c >, we obtain for all, w K M ẽ cm = = ce, w ccα α + inf E, w E c, cw ce, w ccα,w K cm α + α+ dx 86 α+ dx. 87 Let now n, w n be a iniizing sequence of E in K M, such that E n, w n ẽ M + C2M 4 where C 2 M is the constant appearing in 7, which depends on M and on the data of the proble. Recalling that V 0 and L 0, and using 7, we get, for all sufficiently sall, α + α+ n dx 3C 2 R 4 + K 2 α α C 2 M 2 So, this estiate in particular holds for a iniizer of E. Therefore in 87 we get, taking, w to be a iniizer of E which exists by Corollary 4. ẽ cm E c, cw ce, w ccα α+ dx cẽ M cc α C 2M. α + R 2 Hence, we conclude that > 0. ẽ cm cẽ M cc α C 2 2 < cẽ M. 88 Using 88 twice with c = M/a and c = a/m a yields if a > M/2, otherwise it suffices to replace a with M a ẽ M < M a ẽa M a [ α ] M C2 a a 2 = ẽ a + M a ẽ a M a a < ẽ a + ẽ M a M a [ α ] M C2 a a 2 α ] C2 a. 2 [ M a Theore 4.6. Let ũ, λ, and ū, λ, be as in Proposition 4.3. Then, R n dx = M. 89 Proof. Let c > 0 be such that ce x such c exists by Reark 4.4. For R sufficiently large to be chosen later, we define { ce R x R ν R x = ce x x > R. 90 So in particular x ν R x for x > R. 24

25 We observe that as R + ν R xdx R n = Cω e R + Ce x dx 0. \B R 9 Since and H ũ H 0 ū locally uniforly, there exists 0 = 0 R such that for all 0, We observe that for all 0, + H ũ H 0 ū ce R x R ν R ν R x x. 93 Indeed, if x > R, then + 2ν R + ν R ν R, since ν R. On the other hand, if x R, then by ν R ce R + 2 ce R = ce R = ν R. Fro 93 we deduce that + 2ν R. 94 Moreover, since a.e. by Theore 2.3, recalling that dx = M, = a and R n using 9 and 94, we have that + 2ν R dx = M a + 2 ν R dx M a as R +, 95 R li α+ dx = α+ dx + li α+ dx R α+ dx + li + 2ν R α+ dx. 0 We consider the function, w L w. This is a convex function in, w. copute w L w = L w, so in particular by 66 we get We w C L γ C L α w L w C L w γ + C L α. 97 Moreover, L w = L w + w L w, therefore, again by 66 we get w C L γ C L α L w α We clai that, if we define w = H 0 ū, then C L w γ + C L α α. 98 E, w E, w + E + 2ν R, w w + 2 ν R + o + o R, 99 where o is an error such that li 0 o = 0. ote that V y+y dx = V y+y dx+ V y+y +2ν R dx 2 V y+y ν R dx. Recalling 65, the estiate 76 and the definition of ν R, we have 2 V y + y ν R dx CR nˆb+ e R. Hence we obtain V y+y dx V y+y dx+ V y+y +2ν R dx CR nˆb+ e R

26 By 96 we get α+ dx α + R α + α+ dx α + + 2ν R α+ dx + o 0 Finally, we estiate the kinetic ters in the energy. Splitting L w dx = L w dx + B R \B R L w dx, we proceed by estiating separately the two ters. Estiates in \ B R. First of all, note that by 82, 67 and 66, we get that L w = L H 0 ū C for coe constant C > 0, just depending on the data. Moreover, recalling that ce x, we get that, eventually enlarging C, \B R L By convexity of the function, w L w, we get that \B R L w + \B R + w \B R dx w dx C e x dx = o R. 02 x >R + 2ν R L \B R + 2ν R L w w + 2 ν R + 2ν R [ + 2ν R L w w + 2 ν R + 2ν R w w + 2 ν R + 2ν R dx 2ν R dx 03 ] w 2 ν R dx. 04 We recall that w = H 0 ū C by 82 and ν R Cν R by definition. Moreover, by 75 and 67, w = H ũ C [ + x nˆb γ ] γ C + x nˆb. Using the triangular inequality we get the following, where the constant C can change fro line to line, w w + 2 ν R + 2ν R H ũ + H 0 ū Cν R ν R + 2ν R + 2ν R C + x nˆb + 2ν R + C + 2ν R + Cν R C + x nˆb + 2ν R on \ B R 0, where we used respectively the fact that + 2ν R, ν R and that + 2ν R ν R. ow, using 98 and 05, we can estiate 03, and by 97 and 05 we can estiate 04. Indeed, we get + 2ν R L w w + 2 ν R 2ν R dx C + x nˆbν R xdx + 2ν R \B R \B R and \B R w [ + 2ν R L w ] w + 2 ν R w +2 ν R dx C + 2ν R + x nˆb γ νr xdx, \B R 26

27 because w C on. Therefore, we ay conclude, possibly enlarging C, that \B R L w dx + 2ν R L \B R w w + 2 ν R dx C + x nˆb γ νr xdx ν R \B R Finally, putting together 02 and 06, we have, choosing C suffficiently large \B R L w dx + + 2ν R L \B R \B R L w dx w w + 2 ν R + 2ν R dx C + x nˆb γ νr xdx. 07 \B R Estiates in B R. Again by convexity of the function, w L w, we get that B R L w + dx B R B R L L w w dx [ dx + w L w ] w w dx. 08 B R We now estiate 08. We recall that w H 0 ū K and also H ũ K by 82, for all 0 R. Then, using this fact and 97 and 98 and recalling 92, we get w L dx = L H 0 ū dx Ce R B R and B R B R w [ L H 0 ū] H u + H ũ H 0 ū dx Ce R. This iplies that for all 0 R L B R ow we observe that by 66, B R +2ν R L w dx B R L w [ w + 2 ν R dx C + 2ν R B R w dx Ce R. 09 w w + 2 ν R + 2ν R ] γ + +2ν R dx. By 94 we get that + 2ν R + 2ν R Ce R, eventually enalarging C. Moreover, reasoning as in 05, we get w w + 2 ν R + 2ν R H ũ + H ũ H 0 ū C + 2ν R + 2ν R where we used that ν R = 0 for x < R, that H ũ K, that by 94 H ũ H 0 ū +2ν R C by 93 and 92. So, we conclude that B R + 2ν R L +2ν R, w w + 2 ν R dx Ce R ν R 27

28 Putting together 09 and 0 we get, choosing C suffficiently large and for all 0 R, B R L w dx L w dx B R + + 2ν R L B R Therefore, suing up, 07, 00 and 0, we conclude w w + 2 ν R dx C e R. + 2ν R E, w E, w + E + 2ν R, w w + 2 ν R 2 +o CR nˆb+ e R C e R C + x nˆb γ e x dx. x >R M a Let now c R = M a+2 ν R dx. We have that c R as R + and c R <. In particular, c R + 2ν R, c R w w + 2 ν R K M a. By the coputation 86 we get c R E + 2ν R, w w + 2 ν R 3 c α R = E c R + 2ν R, c R w w + 2 ν R + c R + 2ν R α+ dx. α + Observe that by 74 there exists C independent of such that 0 + 2ν R α+ dx α+ + α+ + 2ν R α+ α+ C. Therefore, 3 reads E c R + 2ν R, c R w w + 2 ν R Cc R c α R α + c RE + 2ν R, w w + 2 ν R. Using this inequality, and recalling that E, w = ẽ M, we obtain fro 2 ẽ M ẽ a + ẽ M a + c R E + 2ν R, w w + 2 ν R 4 c α R +Cc R α + + o CR nˆb+ e R C e R C + x nˆb γ e x dx. By Lea 4.5, we get that x >R ẽ M ẽ a + ẽ M a Ca, M, where Ca, M > 0 for a < M and CM, M = 0. This iplies in particular that 0 > Ca, M c R E + 2ν R, w w + 2 ν R + C cα R α + + o CR nˆb+ e R C e R C + x nˆb γ e x dx. x >R By 7 we get that there exist K = KM a > 0 such that E + 2ν R, w w + 2 ν R K. So we obtain 0 > Ca, M c R K + C cα R α + +o CR nˆb+ e R C e R C + x nˆb γ e x dx. x >R If a < M, this gives a contradiction, choosing R sufficiently large and < 0 R. 28

FAST DYNAMO ON THE REAL LINE

FAST DYNAMO ON THE REAL LINE FAST DYAMO O THE REAL LIE O. KOZLOVSKI & P. VYTOVA Abstract. In this paper we show that a piecewise expanding ap on the interval, extended to the real line by a non-expanding ap satisfying soe ild hypthesis

More information

3.8 Three Types of Convergence

3.8 Three Types of Convergence 3.8 Three Types of Convergence 3.8 Three Types of Convergence 93 Suppose that we are given a sequence functions {f k } k N on a set X and another function f on X. What does it ean for f k to converge to

More information

PRÜFER SUBSTITUTIONS ON A COUPLED SYSTEM INVOLVING THE p-laplacian

PRÜFER SUBSTITUTIONS ON A COUPLED SYSTEM INVOLVING THE p-laplacian Electronic Journal of Differential Equations, Vol. 23 (23), No. 23, pp. 9. ISSN: 72-669. URL: http://ejde.ath.txstate.edu or http://ejde.ath.unt.edu ftp ejde.ath.txstate.edu PRÜFER SUBSTITUTIONS ON A COUPLED

More information

Generalized eigenfunctions and a Borel Theorem on the Sierpinski Gasket.

Generalized eigenfunctions and a Borel Theorem on the Sierpinski Gasket. Generalized eigenfunctions and a Borel Theore on the Sierpinski Gasket. Kasso A. Okoudjou, Luke G. Rogers, and Robert S. Strichartz May 26, 2006 1 Introduction There is a well developed theory (see [5,

More information

Computational and Statistical Learning Theory

Computational and Statistical Learning Theory Coputational and Statistical Learning Theory Proble sets 5 and 6 Due: Noveber th Please send your solutions to learning-subissions@ttic.edu Notations/Definitions Recall the definition of saple based Radeacher

More information

2 Q 10. Likewise, in case of multiple particles, the corresponding density in 2 must be averaged over all

2 Q 10. Likewise, in case of multiple particles, the corresponding density in 2 must be averaged over all Lecture 6 Introduction to kinetic theory of plasa waves Introduction to kinetic theory So far we have been odeling plasa dynaics using fluid equations. The assuption has been that the pressure can be either

More information

Distributed Subgradient Methods for Multi-agent Optimization

Distributed Subgradient Methods for Multi-agent Optimization 1 Distributed Subgradient Methods for Multi-agent Optiization Angelia Nedić and Asuan Ozdaglar October 29, 2007 Abstract We study a distributed coputation odel for optiizing a su of convex objective functions

More information

A Bernstein-Markov Theorem for Normed Spaces

A Bernstein-Markov Theorem for Normed Spaces A Bernstein-Markov Theore for Nored Spaces Lawrence A. Harris Departent of Matheatics, University of Kentucky Lexington, Kentucky 40506-0027 Abstract Let X and Y be real nored linear spaces and let φ :

More information

The Weierstrass Approximation Theorem

The Weierstrass Approximation Theorem 36 The Weierstrass Approxiation Theore Recall that the fundaental idea underlying the construction of the real nubers is approxiation by the sipler rational nubers. Firstly, nubers are often deterined

More information

Course Notes for EE227C (Spring 2018): Convex Optimization and Approximation

Course Notes for EE227C (Spring 2018): Convex Optimization and Approximation Course Notes for EE7C (Spring 018: Convex Optiization and Approxiation Instructor: Moritz Hardt Eail: hardt+ee7c@berkeley.edu Graduate Instructor: Max Sichowitz Eail: sichow+ee7c@berkeley.edu October 15,

More information

On the Navier Stokes equations

On the Navier Stokes equations On the Navier Stokes equations Daniel Thoas Hayes April 26, 2018 The proble on the existence and soothness of the Navier Stokes equations is resolved. 1. Proble description The Navier Stokes equations

More information

Asymptotics of weighted random sums

Asymptotics of weighted random sums Asyptotics of weighted rando sus José Manuel Corcuera, David Nualart, Mark Podolskij arxiv:402.44v [ath.pr] 6 Feb 204 February 7, 204 Abstract In this paper we study the asyptotic behaviour of weighted

More information

which together show that the Lax-Milgram lemma can be applied. (c) We have the basic Galerkin orthogonality

which together show that the Lax-Milgram lemma can be applied. (c) We have the basic Galerkin orthogonality UPPSALA UNIVERSITY Departent of Inforation Technology Division of Scientific Coputing Solutions to exa in Finite eleent ethods II 14-6-5 Stefan Engblo, Daniel Elfverson Question 1 Note: a inus sign in

More information

arxiv: v1 [math.pr] 17 May 2009

arxiv: v1 [math.pr] 17 May 2009 A strong law of large nubers for artingale arrays Yves F. Atchadé arxiv:0905.2761v1 [ath.pr] 17 May 2009 March 2009 Abstract: We prove a artingale triangular array generalization of the Chow-Birnbau- Marshall

More information

Constrained Consensus and Optimization in Multi-Agent Networks arxiv: v2 [math.oc] 17 Dec 2008

Constrained Consensus and Optimization in Multi-Agent Networks arxiv: v2 [math.oc] 17 Dec 2008 LIDS Report 2779 1 Constrained Consensus and Optiization in Multi-Agent Networks arxiv:0802.3922v2 [ath.oc] 17 Dec 2008 Angelia Nedić, Asuan Ozdaglar, and Pablo A. Parrilo February 15, 2013 Abstract We

More information

On the Existence of Pure Nash Equilibria in Weighted Congestion Games

On the Existence of Pure Nash Equilibria in Weighted Congestion Games MATHEMATICS OF OPERATIONS RESEARCH Vol. 37, No. 3, August 2012, pp. 419 436 ISSN 0364-765X (print) ISSN 1526-5471 (online) http://dx.doi.org/10.1287/oor.1120.0543 2012 INFORMS On the Existence of Pure

More information

1 Bounding the Margin

1 Bounding the Margin COS 511: Theoretical Machine Learning Lecturer: Rob Schapire Lecture #12 Scribe: Jian Min Si March 14, 2013 1 Bounding the Margin We are continuing the proof of a bound on the generalization error of AdaBoost

More information

Supplementary Material for Fast and Provable Algorithms for Spectrally Sparse Signal Reconstruction via Low-Rank Hankel Matrix Completion

Supplementary Material for Fast and Provable Algorithms for Spectrally Sparse Signal Reconstruction via Low-Rank Hankel Matrix Completion Suppleentary Material for Fast and Provable Algoriths for Spectrally Sparse Signal Reconstruction via Low-Ran Hanel Matrix Copletion Jian-Feng Cai Tianing Wang Ke Wei March 1, 017 Abstract We establish

More information

A := A i : {A i } S. is an algebra. The same object is obtained when the union in required to be disjoint.

A := A i : {A i } S. is an algebra. The same object is obtained when the union in required to be disjoint. 59 6. ABSTRACT MEASURE THEORY Having developed the Lebesgue integral with respect to the general easures, we now have a general concept with few specific exaples to actually test it on. Indeed, so far

More information

Tail estimates for norms of sums of log-concave random vectors

Tail estimates for norms of sums of log-concave random vectors Tail estiates for nors of sus of log-concave rando vectors Rados law Adaczak Rafa l Lata la Alexander E. Litvak Alain Pajor Nicole Toczak-Jaegerann Abstract We establish new tail estiates for order statistics

More information

A Simple Regression Problem

A Simple Regression Problem A Siple Regression Proble R. M. Castro March 23, 2 In this brief note a siple regression proble will be introduced, illustrating clearly the bias-variance tradeoff. Let Y i f(x i ) + W i, i,..., n, where

More information

Computable Shell Decomposition Bounds

Computable Shell Decomposition Bounds Coputable Shell Decoposition Bounds John Langford TTI-Chicago jcl@cs.cu.edu David McAllester TTI-Chicago dac@autoreason.co Editor: Leslie Pack Kaelbling and David Cohn Abstract Haussler, Kearns, Seung

More information

Pacific Journal of Mathematics

Pacific Journal of Mathematics Pacific Journal of Matheatics SINGULAR LIMIT OF SOLUTIONS ( ) OF THE EQUATION u t = u AS 0 Kin Ming Hui Volue 187 No. 2 February 1999 PACIFIC JOURNAL OF MATHEMATICS Vol. 187, No. 2, 1999 SINGULAR LIMIT

More information

Max-Product Shepard Approximation Operators

Max-Product Shepard Approximation Operators Max-Product Shepard Approxiation Operators Barnabás Bede 1, Hajie Nobuhara 2, János Fodor 3, Kaoru Hirota 2 1 Departent of Mechanical and Syste Engineering, Bánki Donát Faculty of Mechanical Engineering,

More information

Solutions of some selected problems of Homework 4

Solutions of some selected problems of Homework 4 Solutions of soe selected probles of Hoework 4 Sangchul Lee May 7, 2018 Proble 1 Let there be light A professor has two light bulbs in his garage. When both are burned out, they are replaced, and the next

More information

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.436J/15.085J Fall 2008 Lecture 11 10/15/2008 ABSTRACT INTEGRATION I

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.436J/15.085J Fall 2008 Lecture 11 10/15/2008 ABSTRACT INTEGRATION I MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.436J/15.085J Fall 2008 Lecture 11 10/15/2008 ABSTRACT INTEGRATION I Contents 1. Preliinaries 2. The ain result 3. The Rieann integral 4. The integral of a nonnegative

More information

Quantum algorithms (CO 781, Winter 2008) Prof. Andrew Childs, University of Waterloo LECTURE 15: Unstructured search and spatial search

Quantum algorithms (CO 781, Winter 2008) Prof. Andrew Childs, University of Waterloo LECTURE 15: Unstructured search and spatial search Quantu algoriths (CO 781, Winter 2008) Prof Andrew Childs, University of Waterloo LECTURE 15: Unstructured search and spatial search ow we begin to discuss applications of quantu walks to search algoriths

More information

On Conditions for Linearity of Optimal Estimation

On Conditions for Linearity of Optimal Estimation On Conditions for Linearity of Optial Estiation Erah Akyol, Kuar Viswanatha and Kenneth Rose {eakyol, kuar, rose}@ece.ucsb.edu Departent of Electrical and Coputer Engineering University of California at

More information

Supplement to: Subsampling Methods for Persistent Homology

Supplement to: Subsampling Methods for Persistent Homology Suppleent to: Subsapling Methods for Persistent Hoology A. Technical results In this section, we present soe technical results that will be used to prove the ain theores. First, we expand the notation

More information

LORENTZ SPACES AND REAL INTERPOLATION THE KEEL-TAO APPROACH

LORENTZ SPACES AND REAL INTERPOLATION THE KEEL-TAO APPROACH LORENTZ SPACES AND REAL INTERPOLATION THE KEEL-TAO APPROACH GUILLERMO REY. Introduction If an operator T is bounded on two Lebesgue spaces, the theory of coplex interpolation allows us to deduce the boundedness

More information

Chapter 6 1-D Continuous Groups

Chapter 6 1-D Continuous Groups Chapter 6 1-D Continuous Groups Continuous groups consist of group eleents labelled by one or ore continuous variables, say a 1, a 2,, a r, where each variable has a well- defined range. This chapter explores:

More information

Ph 20.3 Numerical Solution of Ordinary Differential Equations

Ph 20.3 Numerical Solution of Ordinary Differential Equations Ph 20.3 Nuerical Solution of Ordinary Differential Equations Due: Week 5 -v20170314- This Assignent So far, your assignents have tried to failiarize you with the hardware and software in the Physics Coputing

More information

On the Communication Complexity of Lipschitzian Optimization for the Coordinated Model of Computation

On the Communication Complexity of Lipschitzian Optimization for the Coordinated Model of Computation journal of coplexity 6, 459473 (2000) doi:0.006jco.2000.0544, available online at http:www.idealibrary.co on On the Counication Coplexity of Lipschitzian Optiization for the Coordinated Model of Coputation

More information

Graphical Models in Local, Asymmetric Multi-Agent Markov Decision Processes

Graphical Models in Local, Asymmetric Multi-Agent Markov Decision Processes Graphical Models in Local, Asyetric Multi-Agent Markov Decision Processes Ditri Dolgov and Edund Durfee Departent of Electrical Engineering and Coputer Science University of Michigan Ann Arbor, MI 48109

More information

e-companion ONLY AVAILABLE IN ELECTRONIC FORM

e-companion ONLY AVAILABLE IN ELECTRONIC FORM OPERATIONS RESEARCH doi 10.1287/opre.1070.0427ec pp. ec1 ec5 e-copanion ONLY AVAILABLE IN ELECTRONIC FORM infors 07 INFORMS Electronic Copanion A Learning Approach for Interactive Marketing to a Custoer

More information

Polygonal Designs: Existence and Construction

Polygonal Designs: Existence and Construction Polygonal Designs: Existence and Construction John Hegean Departent of Matheatics, Stanford University, Stanford, CA 9405 Jeff Langford Departent of Matheatics, Drake University, Des Moines, IA 5011 G

More information

The Methods of Solution for Constrained Nonlinear Programming

The Methods of Solution for Constrained Nonlinear Programming Research Inventy: International Journal Of Engineering And Science Vol.4, Issue 3(March 2014), PP 01-06 Issn (e): 2278-4721, Issn (p):2319-6483, www.researchinventy.co The Methods of Solution for Constrained

More information

13.2 Fully Polynomial Randomized Approximation Scheme for Permanent of Random 0-1 Matrices

13.2 Fully Polynomial Randomized Approximation Scheme for Permanent of Random 0-1 Matrices CS71 Randoness & Coputation Spring 018 Instructor: Alistair Sinclair Lecture 13: February 7 Disclaier: These notes have not been subjected to the usual scrutiny accorded to foral publications. They ay

More information

Fairness via priority scheduling

Fairness via priority scheduling Fairness via priority scheduling Veeraruna Kavitha, N Heachandra and Debayan Das IEOR, IIT Bobay, Mubai, 400076, India vavitha,nh,debayan}@iitbacin Abstract In the context of ulti-agent resource allocation

More information

SOME NONLINEAR DIFFERENTIAL INEQUALITIES AND AN APPLICATION TO HÖLDER CONTINUOUS ALMOST COMPLEX STRUCTURES

SOME NONLINEAR DIFFERENTIAL INEQUALITIES AND AN APPLICATION TO HÖLDER CONTINUOUS ALMOST COMPLEX STRUCTURES SOME NONLINEAR DIFFERENTIAL INEQUALITIES AND AN APPLICATION TO HÖLDER CONTINUOUS ALMOST COMPLEX STRUCTURES ADAM COFFMAN AND YIFEI PAN Abstract. We consider soe second order quasilinear partial differential

More information

A Long-Term Mathematical Model for Mining Industries

A Long-Term Mathematical Model for Mining Industries A Long-Ter Matheatical Model for Mining Industries Yves Achdou, Pierre-Noel Giraud, Jean-Michel Lasry, Pierre-Louis Lions October 7, Abstract A parcionious long ter odel is proposed for a ining industry.

More information

List Scheduling and LPT Oliver Braun (09/05/2017)

List Scheduling and LPT Oliver Braun (09/05/2017) List Scheduling and LPT Oliver Braun (09/05/207) We investigate the classical scheduling proble P ax where a set of n independent jobs has to be processed on 2 parallel and identical processors (achines)

More information

New upper bound for the B-spline basis condition number II. K. Scherer. Institut fur Angewandte Mathematik, Universitat Bonn, Bonn, Germany.

New upper bound for the B-spline basis condition number II. K. Scherer. Institut fur Angewandte Mathematik, Universitat Bonn, Bonn, Germany. New upper bound for the B-spline basis condition nuber II. A proof of de Boor's 2 -conjecture K. Scherer Institut fur Angewandte Matheati, Universitat Bonn, 535 Bonn, Gerany and A. Yu. Shadrin Coputing

More information

Lecture October 23. Scribes: Ruixin Qiang and Alana Shine

Lecture October 23. Scribes: Ruixin Qiang and Alana Shine CSCI699: Topics in Learning and Gae Theory Lecture October 23 Lecturer: Ilias Scribes: Ruixin Qiang and Alana Shine Today s topic is auction with saples. 1 Introduction to auctions Definition 1. In a single

More information

On Lotka-Volterra Evolution Law

On Lotka-Volterra Evolution Law Advanced Studies in Biology, Vol. 3, 0, no. 4, 6 67 On Lota-Volterra Evolution Law Farruh Muhaedov Faculty of Science, International Islaic University Malaysia P.O. Box, 4, 570, Kuantan, Pahang, Malaysia

More information

Chaotic Coupled Map Lattices

Chaotic Coupled Map Lattices Chaotic Coupled Map Lattices Author: Dustin Keys Advisors: Dr. Robert Indik, Dr. Kevin Lin 1 Introduction When a syste of chaotic aps is coupled in a way that allows the to share inforation about each

More information

A NEW CHARACTERIZATION OF THE CR SPHERE AND THE SHARP EIGENVALUE ESTIMATE FOR THE KOHN LAPLACIAN. 1. Introduction

A NEW CHARACTERIZATION OF THE CR SPHERE AND THE SHARP EIGENVALUE ESTIMATE FOR THE KOHN LAPLACIAN. 1. Introduction A NEW CHARACTERIZATION OF THE CR SPHERE AND THE SHARP EIGENVALUE ESTIATE FOR THE KOHN LAPLACIAN SONG-YING LI, DUONG NGOC SON, AND XIAODONG WANG 1. Introduction Let, θ be a strictly pseudoconvex pseudo-heritian

More information

Dan Goreac 1 and Oana-Silvia Serea Introduction

Dan Goreac 1 and Oana-Silvia Serea Introduction ESAIM: PROCEEDINGS AND SURVEYS, Septeber 214, Vol. 45, p. 16-177 J.-S. Dhersin, Editor UNIFORM ASSYMPTOTICS IN THE AVERAGE CONTINUOUS CONTROL OF PIECEWISE DETERMINISTIC MARKOV PROCESSES : VANISHING APPROACH,

More information

Model Fitting. CURM Background Material, Fall 2014 Dr. Doreen De Leon

Model Fitting. CURM Background Material, Fall 2014 Dr. Doreen De Leon Model Fitting CURM Background Material, Fall 014 Dr. Doreen De Leon 1 Introduction Given a set of data points, we often want to fit a selected odel or type to the data (e.g., we suspect an exponential

More information

a a a a a a a m a b a b

a a a a a a a m a b a b Algebra / Trig Final Exa Study Guide (Fall Seester) Moncada/Dunphy Inforation About the Final Exa The final exa is cuulative, covering Appendix A (A.1-A.5) and Chapter 1. All probles will be ultiple choice

More information

Department of Physics Preliminary Exam January 3 6, 2006

Department of Physics Preliminary Exam January 3 6, 2006 Departent of Physics Preliinary Exa January 3 6, 2006 Day 1: Classical Mechanics Tuesday, January 3, 2006 9:00 a.. 12:00 p.. Instructions: 1. Write the answer to each question on a separate sheet of paper.

More information

Numerically repeated support splitting and merging phenomena in a porous media equation with strong absorption. Kenji Tomoeda

Numerically repeated support splitting and merging phenomena in a porous media equation with strong absorption. Kenji Tomoeda Journal of Math-for-Industry, Vol. 3 (C-), pp. Nuerically repeated support splitting and erging phenoena in a porous edia equation with strong absorption To the eory of y friend Professor Nakaki. Kenji

More information

Principles of Optimal Control Spring 2008

Principles of Optimal Control Spring 2008 MIT OpenCourseWare http://ocw.it.edu 16.323 Principles of Optial Control Spring 2008 For inforation about citing these aterials or our Ters of Use, visit: http://ocw.it.edu/ters. 16.323 Lecture 10 Singular

More information

Some Classical Ergodic Theorems

Some Classical Ergodic Theorems Soe Classical Ergodic Theores Matt Rosenzweig Contents Classical Ergodic Theores. Mean Ergodic Theores........................................2 Maxial Ergodic Theore.....................................

More information

Asynchronous Gossip Algorithms for Stochastic Optimization

Asynchronous Gossip Algorithms for Stochastic Optimization Asynchronous Gossip Algoriths for Stochastic Optiization S. Sundhar Ra ECE Dept. University of Illinois Urbana, IL 680 ssrini@illinois.edu A. Nedić IESE Dept. University of Illinois Urbana, IL 680 angelia@illinois.edu

More information

A Note on Scheduling Tall/Small Multiprocessor Tasks with Unit Processing Time to Minimize Maximum Tardiness

A Note on Scheduling Tall/Small Multiprocessor Tasks with Unit Processing Time to Minimize Maximum Tardiness A Note on Scheduling Tall/Sall Multiprocessor Tasks with Unit Processing Tie to Miniize Maxiu Tardiness Philippe Baptiste and Baruch Schieber IBM T.J. Watson Research Center P.O. Box 218, Yorktown Heights,

More information

Sharp Time Data Tradeoffs for Linear Inverse Problems

Sharp Time Data Tradeoffs for Linear Inverse Problems Sharp Tie Data Tradeoffs for Linear Inverse Probles Saet Oyak Benjain Recht Mahdi Soltanolkotabi January 016 Abstract In this paper we characterize sharp tie-data tradeoffs for optiization probles used

More information

ADVANCES ON THE BESSIS- MOUSSA-VILLANI TRACE CONJECTURE

ADVANCES ON THE BESSIS- MOUSSA-VILLANI TRACE CONJECTURE ADVANCES ON THE BESSIS- MOUSSA-VILLANI TRACE CONJECTURE CHRISTOPHER J. HILLAR Abstract. A long-standing conjecture asserts that the polynoial p(t = Tr(A + tb ] has nonnegative coefficients whenever is

More information

Computable Shell Decomposition Bounds

Computable Shell Decomposition Bounds Journal of Machine Learning Research 5 (2004) 529-547 Subitted 1/03; Revised 8/03; Published 5/04 Coputable Shell Decoposition Bounds John Langford David McAllester Toyota Technology Institute at Chicago

More information

ANALYSIS OF A FULLY DISCRETE FINITE ELEMENT METHOD FOR THE PHASE FIELD MODEL AND APPROXIMATION OF ITS SHARP INTERFACE LIMITS

ANALYSIS OF A FULLY DISCRETE FINITE ELEMENT METHOD FOR THE PHASE FIELD MODEL AND APPROXIMATION OF ITS SHARP INTERFACE LIMITS ANALYSIS OF A FULLY DISCRETE FINITE ELEMENT METHOD FOR THE PHASE FIELD MODEL AND APPROXIMATION OF ITS SHARP INTERFACE LIMITS XIAOBING FENG AND ANDREAS PROHL Abstract. We propose and analyze a fully discrete

More information

E0 370 Statistical Learning Theory Lecture 6 (Aug 30, 2011) Margin Analysis

E0 370 Statistical Learning Theory Lecture 6 (Aug 30, 2011) Margin Analysis E0 370 tatistical Learning Theory Lecture 6 (Aug 30, 20) Margin Analysis Lecturer: hivani Agarwal cribe: Narasihan R Introduction In the last few lectures we have seen how to obtain high confidence bounds

More information

Lecture 20 November 7, 2013

Lecture 20 November 7, 2013 CS 229r: Algoriths for Big Data Fall 2013 Prof. Jelani Nelson Lecture 20 Noveber 7, 2013 Scribe: Yun Willia Yu 1 Introduction Today we re going to go through the analysis of atrix copletion. First though,

More information

Curious Bounds for Floor Function Sums

Curious Bounds for Floor Function Sums 1 47 6 11 Journal of Integer Sequences, Vol. 1 (018), Article 18.1.8 Curious Bounds for Floor Function Sus Thotsaporn Thanatipanonda and Elaine Wong 1 Science Division Mahidol University International

More information

Non-Parametric Non-Line-of-Sight Identification 1

Non-Parametric Non-Line-of-Sight Identification 1 Non-Paraetric Non-Line-of-Sight Identification Sinan Gezici, Hisashi Kobayashi and H. Vincent Poor Departent of Electrical Engineering School of Engineering and Applied Science Princeton University, Princeton,

More information

Tight Bounds for Maximal Identifiability of Failure Nodes in Boolean Network Tomography

Tight Bounds for Maximal Identifiability of Failure Nodes in Boolean Network Tomography Tight Bounds for axial Identifiability of Failure Nodes in Boolean Network Toography Nicola Galesi Sapienza Università di Roa nicola.galesi@uniroa1.it Fariba Ranjbar Sapienza Università di Roa fariba.ranjbar@uniroa1.it

More information

TD M1 EDP 2018 no 2 Elliptic equations: regularity, maximum principle

TD M1 EDP 2018 no 2 Elliptic equations: regularity, maximum principle TD M EDP 08 no Elliptic equations: regularity, maximum principle Estimates in the sup-norm I Let be an open bounded subset of R d of class C. Let A = (a ij ) be a symmetric matrix of functions of class

More information

Fixed-to-Variable Length Distribution Matching

Fixed-to-Variable Length Distribution Matching Fixed-to-Variable Length Distribution Matching Rana Ali Ajad and Georg Böcherer Institute for Counications Engineering Technische Universität München, Gerany Eail: raa2463@gail.co,georg.boecherer@tu.de

More information

Tight Information-Theoretic Lower Bounds for Welfare Maximization in Combinatorial Auctions

Tight Information-Theoretic Lower Bounds for Welfare Maximization in Combinatorial Auctions Tight Inforation-Theoretic Lower Bounds for Welfare Maxiization in Cobinatorial Auctions Vahab Mirrokni Jan Vondrák Theory Group, Microsoft Dept of Matheatics Research Princeton University Redond, WA 9805

More information

Uniform Approximation and Bernstein Polynomials with Coefficients in the Unit Interval

Uniform Approximation and Bernstein Polynomials with Coefficients in the Unit Interval Unifor Approxiation and Bernstein Polynoials with Coefficients in the Unit Interval Weiang Qian and Marc D. Riedel Electrical and Coputer Engineering, University of Minnesota 200 Union St. S.E. Minneapolis,

More information

Lost-Sales Problems with Stochastic Lead Times: Convexity Results for Base-Stock Policies

Lost-Sales Problems with Stochastic Lead Times: Convexity Results for Base-Stock Policies OPERATIONS RESEARCH Vol. 52, No. 5, Septeber October 2004, pp. 795 803 issn 0030-364X eissn 1526-5463 04 5205 0795 infors doi 10.1287/opre.1040.0130 2004 INFORMS TECHNICAL NOTE Lost-Sales Probles with

More information

Lower Bounds for Quantized Matrix Completion

Lower Bounds for Quantized Matrix Completion Lower Bounds for Quantized Matrix Copletion Mary Wootters and Yaniv Plan Departent of Matheatics University of Michigan Ann Arbor, MI Eail: wootters, yplan}@uich.edu Mark A. Davenport School of Elec. &

More information

A BLOCK MONOTONE DOMAIN DECOMPOSITION ALGORITHM FOR A NONLINEAR SINGULARLY PERTURBED PARABOLIC PROBLEM

A BLOCK MONOTONE DOMAIN DECOMPOSITION ALGORITHM FOR A NONLINEAR SINGULARLY PERTURBED PARABOLIC PROBLEM INTERNATIONAL JOURNAL OF NUMERICAL ANALYSIS AND MODELING Volue 3, Nuber 2, Pages 211 231 c 2006 Institute for Scientific Coputing and Inforation A BLOCK MONOTONE DOMAIN DECOMPOSITION ALGORITHM FOR A NONLINEAR

More information

PREPRINT 2006:17. Inequalities of the Brunn-Minkowski Type for Gaussian Measures CHRISTER BORELL

PREPRINT 2006:17. Inequalities of the Brunn-Minkowski Type for Gaussian Measures CHRISTER BORELL PREPRINT 2006:7 Inequalities of the Brunn-Minkowski Type for Gaussian Measures CHRISTER BORELL Departent of Matheatical Sciences Division of Matheatics CHALMERS UNIVERSITY OF TECHNOLOGY GÖTEBORG UNIVERSITY

More information

On Constant Power Water-filling

On Constant Power Water-filling On Constant Power Water-filling Wei Yu and John M. Cioffi Electrical Engineering Departent Stanford University, Stanford, CA94305, U.S.A. eails: {weiyu,cioffi}@stanford.edu Abstract This paper derives

More information

The isomorphism problem of Hausdorff measures and Hölder restrictions of functions

The isomorphism problem of Hausdorff measures and Hölder restrictions of functions The isoorphis proble of Hausdorff easures and Hölder restrictions of functions Doctoral thesis András Máthé PhD School of Matheatics Pure Matheatics Progra School Leader: Prof. Miklós Laczkovich Progra

More information

Theore A. Let n (n 4) be arcoplete inial iersed hypersurface in R n+. 4(n ) Then there exists a constant C (n) = (n 2)n S (n) > 0 such that if kak n d

Theore A. Let n (n 4) be arcoplete inial iersed hypersurface in R n+. 4(n ) Then there exists a constant C (n) = (n 2)n S (n) > 0 such that if kak n d Gap theores for inial subanifolds in R n+ Lei Ni Departent of atheatics Purdue University West Lafayette, IN 47907 99 atheatics Subject Classification 53C2 53C42 Introduction Let be a copact iersed inial

More information

Soft Computing Techniques Help Assign Weights to Different Factors in Vulnerability Analysis

Soft Computing Techniques Help Assign Weights to Different Factors in Vulnerability Analysis Soft Coputing Techniques Help Assign Weights to Different Factors in Vulnerability Analysis Beverly Rivera 1,2, Irbis Gallegos 1, and Vladik Kreinovich 2 1 Regional Cyber and Energy Security Center RCES

More information

Course Notes for EE227C (Spring 2018): Convex Optimization and Approximation

Course Notes for EE227C (Spring 2018): Convex Optimization and Approximation Course Notes for EE227C (Spring 2018): Convex Optiization and Approxiation Instructor: Moritz Hardt Eail: hardt+ee227c@berkeley.edu Graduate Instructor: Max Sichowitz Eail: sichow+ee227c@berkeley.edu October

More information

Math Reviews classifications (2000): Primary 54F05; Secondary 54D20, 54D65

Math Reviews classifications (2000): Primary 54F05; Secondary 54D20, 54D65 The Monotone Lindelöf Property and Separability in Ordered Spaces by H. Bennett, Texas Tech University, Lubbock, TX 79409 D. Lutzer, College of Willia and Mary, Williasburg, VA 23187-8795 M. Matveev, Irvine,

More information

E0 370 Statistical Learning Theory Lecture 5 (Aug 25, 2011)

E0 370 Statistical Learning Theory Lecture 5 (Aug 25, 2011) E0 370 Statistical Learning Theory Lecture 5 Aug 5, 0 Covering Nubers, Pseudo-Diension, and Fat-Shattering Diension Lecturer: Shivani Agarwal Scribe: Shivani Agarwal Introduction So far we have seen how

More information

This model assumes that the probability of a gap has size i is proportional to 1/i. i.e., i log m e. j=1. E[gap size] = i P r(i) = N f t.

This model assumes that the probability of a gap has size i is proportional to 1/i. i.e., i log m e. j=1. E[gap size] = i P r(i) = N f t. CS 493: Algoriths for Massive Data Sets Feb 2, 2002 Local Models, Bloo Filter Scribe: Qin Lv Local Models In global odels, every inverted file entry is copressed with the sae odel. This work wells when

More information

Lecture #8-3 Oscillations, Simple Harmonic Motion

Lecture #8-3 Oscillations, Simple Harmonic Motion Lecture #8-3 Oscillations Siple Haronic Motion So far we have considered two basic types of otion: translation and rotation. But these are not the only two types of otion we can observe in every day life.

More information

arxiv: v2 [math.ap] 28 Nov 2016

arxiv: v2 [math.ap] 28 Nov 2016 ONE-DIMENSIONAL SAIONARY MEAN-FIELD GAMES WIH LOCAL COUPLING DIOGO A. GOMES, LEVON NURBEKYAN, AND MARIANA PRAZERES arxiv:1611.8161v [math.ap] 8 Nov 16 Abstract. A standard assumption in mean-field game

More information

A note on the multiplication of sparse matrices

A note on the multiplication of sparse matrices Cent. Eur. J. Cop. Sci. 41) 2014 1-11 DOI: 10.2478/s13537-014-0201-x Central European Journal of Coputer Science A note on the ultiplication of sparse atrices Research Article Keivan Borna 12, Sohrab Aboozarkhani

More information

Lecture 21. Interior Point Methods Setup and Algorithm

Lecture 21. Interior Point Methods Setup and Algorithm Lecture 21 Interior Point Methods In 1984, Kararkar introduced a new weakly polynoial tie algorith for solving LPs [Kar84a], [Kar84b]. His algorith was theoretically faster than the ellipsoid ethod and

More information

4 = (0.02) 3 13, = 0.25 because = 25. Simi-

4 = (0.02) 3 13, = 0.25 because = 25. Simi- Theore. Let b and be integers greater than. If = (. a a 2 a i ) b,then for any t N, in base (b + t), the fraction has the digital representation = (. a a 2 a i ) b+t, where a i = a i + tk i with k i =

More information

SOLUTIONS. PROBLEM 1. The Hamiltonian of the particle in the gravitational field can be written as, x 0, + U(x), U(x) =

SOLUTIONS. PROBLEM 1. The Hamiltonian of the particle in the gravitational field can be written as, x 0, + U(x), U(x) = SOLUTIONS PROBLEM 1. The Hailtonian of the particle in the gravitational field can be written as { Ĥ = ˆp2, x 0, + U(x), U(x) = (1) 2 gx, x > 0. The siplest estiate coes fro the uncertainty relation. If

More information

Deflation of the I-O Series Some Technical Aspects. Giorgio Rampa University of Genoa April 2007

Deflation of the I-O Series Some Technical Aspects. Giorgio Rampa University of Genoa April 2007 Deflation of the I-O Series 1959-2. Soe Technical Aspects Giorgio Rapa University of Genoa g.rapa@unige.it April 27 1. Introduction The nuber of sectors is 42 for the period 1965-2 and 38 for the initial

More information

The path integral approach in the frame work of causal interpretation

The path integral approach in the frame work of causal interpretation Annales de la Fondation Louis de Broglie, Volue 28 no 1, 2003 1 The path integral approach in the frae work of causal interpretation M. Abolhasani 1,2 and M. Golshani 1,2 1 Institute for Studies in Theoretical

More information

On Poset Merging. 1 Introduction. Peter Chen Guoli Ding Steve Seiden. Keywords: Merging, Partial Order, Lower Bounds. AMS Classification: 68W40

On Poset Merging. 1 Introduction. Peter Chen Guoli Ding Steve Seiden. Keywords: Merging, Partial Order, Lower Bounds. AMS Classification: 68W40 On Poset Merging Peter Chen Guoli Ding Steve Seiden Abstract We consider the follow poset erging proble: Let X and Y be two subsets of a partially ordered set S. Given coplete inforation about the ordering

More information

TRAVELING WAVE SOLUTIONS OF THE POROUS MEDIUM EQUATION. Laxmi P. Paudel. Dissertation Prepared for the Degree of DOCTOR OF PHILOSOPHY

TRAVELING WAVE SOLUTIONS OF THE POROUS MEDIUM EQUATION. Laxmi P. Paudel. Dissertation Prepared for the Degree of DOCTOR OF PHILOSOPHY TRAVELING WAVE SOLUTIONS OF THE POROUS MEDIUM EQUATION Laxi P Paudel Dissertation Prepared for the Degree of DOCTOR OF PHILOSOPHY UNIVERSITY OF NORTH TEXAS May 013 APPROVED: Joseph Iaia, Major Professor

More information

Keywords: Estimator, Bias, Mean-squared error, normality, generalized Pareto distribution

Keywords: Estimator, Bias, Mean-squared error, normality, generalized Pareto distribution Testing approxiate norality of an estiator using the estiated MSE and bias with an application to the shape paraeter of the generalized Pareto distribution J. Martin van Zyl Abstract In this work the norality

More information

Kernel Methods and Support Vector Machines

Kernel Methods and Support Vector Machines Intelligent Systes: Reasoning and Recognition Jaes L. Crowley ENSIAG 2 / osig 1 Second Seester 2012/2013 Lesson 20 2 ay 2013 Kernel ethods and Support Vector achines Contents Kernel Functions...2 Quadratic

More information

An Optimal Family of Exponentially Accurate One-Bit Sigma-Delta Quantization Schemes

An Optimal Family of Exponentially Accurate One-Bit Sigma-Delta Quantization Schemes An Optial Faily of Exponentially Accurate One-Bit Siga-Delta Quantization Schees Percy Deift C. Sinan Güntürk Felix Kraher January 2, 200 Abstract Siga-Delta odulation is a popular ethod for analog-to-digital

More information

Feature Extraction Techniques

Feature Extraction Techniques Feature Extraction Techniques Unsupervised Learning II Feature Extraction Unsupervised ethods can also be used to find features which can be useful for categorization. There are unsupervised ethods that

More information

Multi-Dimensional Hegselmann-Krause Dynamics

Multi-Dimensional Hegselmann-Krause Dynamics Multi-Diensional Hegselann-Krause Dynaics A. Nedić Industrial and Enterprise Systes Engineering Dept. University of Illinois Urbana, IL 680 angelia@illinois.edu B. Touri Coordinated Science Laboratory

More information

lecture 36: Linear Multistep Mehods: Zero Stability

lecture 36: Linear Multistep Mehods: Zero Stability 95 lecture 36: Linear Multistep Mehods: Zero Stability 5.6 Linear ultistep ethods: zero stability Does consistency iply convergence for linear ultistep ethods? This is always the case for one-step ethods,

More information

A Better Algorithm For an Ancient Scheduling Problem. David R. Karger Steven J. Phillips Eric Torng. Department of Computer Science

A Better Algorithm For an Ancient Scheduling Problem. David R. Karger Steven J. Phillips Eric Torng. Department of Computer Science A Better Algorith For an Ancient Scheduling Proble David R. Karger Steven J. Phillips Eric Torng Departent of Coputer Science Stanford University Stanford, CA 9435-4 Abstract One of the oldest and siplest

More information

Celal S. Konor Release 1.1 (identical to 1.0) 3/21/08. 1-Hybrid isentropic-sigma vertical coordinate and governing equations in the free atmosphere

Celal S. Konor Release 1.1 (identical to 1.0) 3/21/08. 1-Hybrid isentropic-sigma vertical coordinate and governing equations in the free atmosphere Celal S. Konor Release. (identical to.0) 3/2/08 -Hybrid isentropic-siga vertical coordinate governing equations in the free atosphere This section describes the equations in the free atosphere of the odel.

More information

Learnability and Stability in the General Learning Setting

Learnability and Stability in the General Learning Setting Learnability and Stability in the General Learning Setting Shai Shalev-Shwartz TTI-Chicago shai@tti-c.org Ohad Shair The Hebrew University ohadsh@cs.huji.ac.il Nathan Srebro TTI-Chicago nati@uchicago.edu

More information