• Aucun résultat trouvé

Asymptotic derivation of the section-averaged shallow water equations for natural river hydraulics.

N/A
N/A
Protected

Academic year: 2021

Partager "Asymptotic derivation of the section-averaged shallow water equations for natural river hydraulics."

Copied!
31
0
0

Texte intégral

(1)

HAL Id: hal-00275460

https://hal.archives-ouvertes.fr/hal-00275460v2

Preprint submitted on 8 Jul 2008

HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

water equations for natural river hydraulics.

Astrid Decoene, Luca Bonaventura, Edie Miglio, Fausto Saleri

To cite this version:

Astrid Decoene, Luca Bonaventura, Edie Miglio, Fausto Saleri. Asymptotic derivation of the section- averaged shallow water equations for natural river hydraulics.. 2008. �hal-00275460v2�

(2)

water equations for natural river hydraulis

Astrid Deoene

, Lua Bonaventura

, Edie Miglio

,Fausto Saleri

MOX Modelling and Sienti Computing

Dipartimentodi Matematia F. Brioshi,Politenio di Milano

Via Bonardi 9, 20133 Milano, Italy

Abstrat

The setion-averagedshallow water model usually applied in river and open hannel

hydraulisisderivedasymptotiallyuptoseondorder inthevertial/longitudinallength

ratio,starting from the three-dimensionalReynolds-averagedNavier-Stokesequations for

inompressible free surfae ows. The derivation is arried out under quite general as-

sumptionsonthegeometry ofthehannel, thusallowingfortheappliationof theresult-

ingequations to naturalriverswith arbitrarily shaped ross setions. As a resultof the

derivation, ageneralizedfrition termis obtained, that does notrelyon loal uniformity

assumptionsandthatanbeomputeddiretlyfromthree-dimensionalturbulenemodels,

withoutneedforloaluniformityassumptions. Themodiedequationsinludingthenovel

frition term are ompared to the lassialSaint Venant equations in the ase of steady

stateopenhannelows,whereanalytisolutionsareavailable,showingthatthesolutions

resultingfromthemodiedequationsetaremuhlosertothethree-dimensionalsolutions

thanthoseofthelassialequationset. Furthermore,itisshownthattheproposedformu-

lationyieldsresultsthatareverysimilartothoseobtainedwithempirialfritionlosures

widelyappliedinomputationalhydraulis. Thegeneralizedfritiontermderivedtherefore

justiesaposteriori these empiriallosures, while allowingto avoidthe assumptions on

loalowuniformityonwhihtheselosuresrely.

Keywords: Computational hydraulis; shallow water equations;Saint Venant equations.

1 Introdution

In environmental modelling of freesurfae ows, wheneverthe ratio between the vertial and

longitudinalsalesissmallenough,theso-alled Shallow Water approximationisusuallyintro-

dued, in order to redue the omputational ost implied by the numerial solution of three-

dimensional free surfae ow equations. Models based on this approximation are extensively

used to simulate various geophysial phenomena,suh as rivers and oastal ows,[6 ,8℄ oeans

and even avalanhes,[1℄ and they have been used in hydraulis for a very long time. When

the visosity is negleted and a retangular hannel setion is assumed, the derivation of the

astrid.deoenemate.polimi.it

lua.bonaventurapolimi.it

edie.migliomate.polimi.it

(3)

one-dimensionalShallow Watersystemis lassial,seee.g.[13 ℄. However, thisderivation isun-

satisfatory, sine visosity eets areadded a posteriori and thethree-dimensional geometry

isnot arbitrary.

In Ref.[7℄, Gerbeau and Perthame derive rigourously, by asymptoti analysis, a one -

dimensional visous Saint-Venant system from the two-dimensional Navier-Stokes equations

withmoleularvisosity,linearfritionboundary onditionsand atbathymetry. Theeet of

thevisosityisreovered inaone-dimensionalfrition termand inaone-dimensional diusion

term, both resulting from the derivation. The nal system is a seond order approximation

with respet to the ratio between the vertial and longitudinal sales of the original two-

dimensionalmodel. Othersystemshavebeenderivedinthesamespirit. InRef.[12℄,theasymp-

totianalysisismadethroughavariablehangeinareferenedomain,independentoftheratio

parameter and time. Marhe proposes inRef.[10℄ the derivation of a two-dimensional visous

shallowwatersystemtakingintoaountapillaryeets, varyingbathymetryandamoleular

visosity. However, inorder to simulate realisti river ows,three-dimensional geometries and

turbulenephenomenamustbetakenintoaount. Thus,theReynolds-averagedNavier-Stokes

equations (RANS) on an arbitrary three-dimensional domain are a more appropriate starting

pointforthe derivation ofsimpliedsystems. InRef.[5℄,Salerietal.derivedatwo-dimensional

visousshallowwatersystemfromthethree-dimensional RANSequations,takinginto aount

a non-at bathymetry, atmospheri pressure eets and onsidering a onstant vertial eddy

visosityandlinear fritionboundaryonditions.

In this paper, we have hosento proeedas inRef.[7℄, extendingthe analysisto thethree-

dimensionalRANSequationswithanisotropiReynoldstensorforfreesurfaeowsinarbitrary

geometry,withnonlinearfritionboundaryonditionsentirely analogoustothoseatuallyused

inpratieinhydraulisappliations. Wepresentarigourousderivationofthesetion-averaged

system, inluding the eets of eddy visosity and frition. This derivation is also aimed at

providing an adequate framework for the rigorous derivation of oupling between three- and

one-dimensionalfreesurfae models. Theequationsystemobtained allowsto omputethefree

surfae level of the ow as well as a setion-averaged veloity. If applied to hannels with

retangular ross-setion, this systemis similar to thelassial setion-averaged shallow water

equations [11℄,exeptforthefrition term. Indeed,our derivation showsthat, inorderto take

into aount eets up to theseond order inthe asymptoti parameter, the lassial frition

term should be orreted by a term whih depends on the turbulent vertial visosity. This

onlusion is in agreement withthe one of Gerbeau etal. in Ref.[7℄ for two-dimensional ows

with onstant visosity over a at bathymetry. Indeed, if the vertial visosity and frition

oeients aretaken to beonstant and the ow is homogeneousin thetransversal diretion,

we retrieve the same frition orretion as in Ref.[7℄. However, our derivation provides the

expression of the frition orretion term in a more general ase, whih inludes turbulent

ows,nonlinearfrition boundary onditions and three-dimensional arbitrarygeometry.

In partiular, we ompute the orretion term assoiated to a spei model for the verti-

al prole of turbulent veloity. Furthermore, for steady state open hannel ows admitting

analyti solutions of the three-dimensional aswell asthe simplied models, we show that the

solutions omputed inludingour orretion term aremuhloser tothose of thethree dimen-

sional model than those of the standard shallow water model. If empirial frition losures

areintrodued,asommonlydone inomputationalhydraulis(seee.g.[2 ℄), oneobtains results

very similar to those of our generalized frition term for steady state open hannel solutions.

Thus, the generalized frition term resulting from the present derivation justies a posteriori

these empirial losures, while allowing to avoid the assumptions on loal ow uniformity on

whih these losures rely. The frition orretion term an be easily inluded in setion aver-

(4)

aged modelssuh asthe one proposed byDeponti etal. inRef.[3℄, [4℄. Its useisalso expeted

to easethe oupling of three- and one-dimensional freesurfae models in theframework of an

integrated hydrologial basin model.

Insetion2ofthispaperwereviewthethree-dimensionalRANSequationsandtheirbound-

ary onditions. Then, we derive the setion-averaged shallow water modelin setion 3 and in

setion 4 we give the expression of the frition orretion term in the laminar and turbulent

ases. Finally, in setion 5, we ompare the analytial solutions of thethree-dimensional and

the setion-averaged models in the partiular ase of steady state open hannel ows with

retangular ross-setion, in order to show the auray gain ahieved by adding the frition

orretion.

2 The Reynolds-averaged Navier-Stokes equations

2.1 The three-dimensional equations with boundary onditions

We onsider the motion of an inompressible uid with onstant density ρ >0, in a three-

dimensional domaint = Ω(t) whih is normal with respet to the vertial diretion z. We

denote by ω the projetion oft on thexy-plane,dened asfollows:

ω(t) =

(x, y)∈R2/0≤x≤L, l1(x, t)≤y≤l2(x, t) ,

where l1 and l2 are the time and spae dependent transversal limits of the ow, and L its

length. We assumethebottom of thedomain to be xedand impervious. We allη andb the

funtionsdesribingthefreesurfaeandthebottom,whihisassumedtobexedintime. The

three-dimensional domainis dened by

t=

(x, y, z)∈R3/(x, y)∈ω(t), b(x, y)≤z≤η(x, y, t) ,

as illustrated in Figure 1. The boundary of the domaint is denoted by ∂Ωt and an be

deomposed into four separate parts: the freesurfae Γs(t),the bottom surfae Γb,the inow

boundary Γin(t) andtheoutow boundary Γout(t).

Figure 1: Three-dimensional domain

The governing equations for the motion of the uid are the inompressible Reynolds-

AveragedNavier-Stokes(RANS)equations int,valid foranyt∈(0, T],whihan be written

(5)

asfollows:

 dU

dt −∇· 1

ρσT

= f +g,

∇·U = 0,

(1)

where U = (u, v, w)T is the veloity of the uid, σT is the stress tensor, f = (fx, fy, fz)T

the sum of the external fores applied on the uid,

d

dt denotes the total time derivative and g = (0,0,−g)T the gravity aeleration. We only onsider Newtonian uids, for whih the tensor σT is written inthe following way:

σT = −pI + σ, (2)

where p is the pressure and σ the visous stress tensor. Sine we are onsidering ows in

preseneof gravity, thatis aligned withthevertial diretion, we onsidera turbulene model

giventhroughananisotropirelationshipbetweenthestresstensorσ andthestrain-ratetensor

D=∇U + (∇U)T.

Following Levermore andSammartino inRef. [9℄,we take:

σ =

σ11 µhD12 µvD13 µhD21 σ22 µvD23 µvD31 µvD32 µeD33

, (3)

where

σ11h(D11−1

2(D11+D22)) +µe1

2(D11+D22),

and

σ22h(D22−1

2(D11+D22)) +µe1

2(D11+D22).

Thepositiveoeientsµh,µv andµe aretheeddyvisosities. Theyan beinterpreted asthe eddyvisosityrelativetothe horizontalshearmotion,theeddyvisosityrelativetothevertial

shear motion, andthe bulkvisosityrelative to the expansionrate inthehorizontal diretion,

respetively.

The system is losed by suitable initial and boundary onditions. We denote by ns the

outward normalto the freesurfae, whihdependson time:

ns= 1

p1 +|∇η|2 (−∂η

∂x,−∂η

∂y,1)T,

and bynb theoutward normalto thebottom:

nb = 1

p1 +|∇b|2(∂b

∂x,∂b

∂y,−1)T,

whilevetors

tb,1

and

tb,2

forma basisfor the tangent planeto thebottom surfae:

tb,1= 1 s

1 +

∂b

∂x

2 (1,0,∂b

∂x)T,

(6)

and

tb,2= 1 s

1 +

∂b

∂y

2(0,1,∂b

∂y)T.

In a visous ow, the veloity is zero on a solidwall, sothat the so-alled no-slip ondition

should beapplied on thebottom:

U = 0 on Γb. (4)

However, the boundary layer at the bottom is hardly ever resolved at typial resolutions of

environmentalmodels. Furthermore,itisneessarytodesribeinsomeapproximatefashionthe

subgrid sale surfae roughness. Thus, ondition (4)is generallysubstituted by two boundary

onditions assignedat a small distane∆zr from thewall, whihrepresents thetypial length

sale of the bottom boundary layer. In addition, the veloity is onsidered zero at a distane

∆z0 ofthe wall,whihrepresentsthe typiallength saleofthebottomsurfae roughness,and

should bemuh smallerthan ∆zr:

∆z0 <<∆zr. (5)

The rstboundary onditionis akinemati ondition:

U ·nb = 0 at z=b(x, y) + ∆zr, (6)

and theseond oneis adynami onditionwhih aounts forfrition eets:

(1

ρσT ·nb)T ·tb = −α||U||U·tb at z=b(x, y) + ∆zr, (7)

where α > 0 is a dimensionless frition oeient. Note that nb and tb are, respetively, the outward normalanda tangent vetorto thebottom surfae,sothatondition (7)isindeedan

assumption on the prole of the tangential veloity omponent along the diretion normal to

the bottom surfae. A logarithmi wall lawis usually assumedfor tangential veloity nearthe

bottom,seee.g. [11 ℄,sothataparaboli modelishosenfor thevertial eddyvisosity,aswell

as a partiular value of the frition oeient α, depending on the value of ∆zr,that will be

desribedingreater detaillater. Inthefollowing we will denotezr(x, y) =b(x, y) + ∆zr.

Atthefreesurfae,theveloityoftheuidisequal totheveloityof thefreesurfae itself.

This isexpressedbythe following kinemati ondition:

∂η

∂t − U ·ns = 0 on Γs(t). (8)

The dynamialondition atthefree surfaetakesinto aount theatmospheri stress,

1

ρσT·ns = −1

ρpans on Γs(t), (9)

where pa isthe atmospheripressure.

(7)

2.2 Adimensionalization of the system

Let us onsider the following absolute sales: L for the total length, H for the depth and U

for the x-omponent of the veloity. We denote by ǫ the ratio between the vertial and the

longitudinalsales:

ǫ= H L.

Inaddition we introdue the following dimensionlessquantities:

νh = µh

ρ U L, νv = µv

ρ U L, νe= µe

ρ U L, G= H

U2g, pa= pa ρ U2.

ThesalefortimeisL/U,forthevertialveloityitisW =ǫ U,andforthepressureP =ρ U2.

For the sake of simpliity we indiate again by u, v, w, p, η and b, respetively, veloity omponents, pressure,freesurfae andbottomelevation, afterresaling. Usingthese notations

in(1) we obtain the following adimensionalized system, written asa funtion of the primitive

unknowns u,v,wand p:

















































































∂u

∂t +∂u2

∂x +∂uv

∂y + ∂uw

∂z + ∂p

∂x = ∂

∂x

he)∂u

∂x −(νh−νe)∂v

∂y

+ ∂

∂y

νh(∂u

∂y +∂v

∂x)

+ 1 ǫ2

∂z

νv∂u

∂z

+ ∂

∂z

νv∂w

∂x

,

∂v

∂t +∂uv

∂x +∂v2

∂y +∂vw

∂z +∂p

∂y = ∂

∂x

νh(∂u

∂y+ ∂v

∂x)

+ ∂

∂y

he)∂v

∂y −(νh−νe)∂u

∂x

+ 1 ǫ2

∂z

νv∂v

∂z

+ ∂

∂z

νv∂w

∂y

,

ǫ2 ∂w

∂t +∂uw

∂x +∂vw

∂y + ∂w2

∂z

+∂p

∂z = −G+ ∂

∂x

νv∂u

∂z +ǫ2νv∂w

∂x

+ ∂

∂y

νv∂v

∂z

2

∂y

νv∂w

∂y

+ ∂

∂z

e∂w

∂z

,

∂u

∂x + ∂v

∂y + ∂w

∂z = 0.

(10)

(8)

Coherently,the resaled boundary onditions are, onthefreesurfae Γs(t),

































































∂η

∂t +u∂η

∂x+v∂η

∂y = w,

∂η

∂x

p−(νhe)∂u

∂x + (νh−νe)∂v

∂y

−∂η

∂y

νh(∂v

∂x+∂u

∂y

v 1

ǫ2

∂u

∂z +∂w

∂x

= pa∂η

∂x,

−∂η

∂x

νh ∂v

∂x +∂u

∂y

+∂η

∂y

p−(νhe)∂v

∂y+ (νh−νe)∂u

∂x

v 1

ǫ2

∂v

∂z +∂w

∂y

= pa∂η

∂y,

−∂η

∂x

νv ∂u

∂z +ǫ2∂w

∂x

−∂η

∂y

νv ∂v

∂z +ǫ2∂w

∂y

−p + 2νe∂w

∂z = −pa,

(11)

and nearthebottom at z=zr,























































































 u∂b

∂x+v∂b

∂y =w,

∂b

∂x

he)∂u

∂x−(νh−νe)∂v

∂y−2νe∂w

∂z

+ ∂b

∂y

νh ∂v

∂x +∂u

∂y

+

∂b

∂x 2

− 1 ǫ2

! νv

∂u

∂z +ǫ2∂w

∂x

+ ∂b

∂x

∂b

∂y

νv ∂v

∂z +ǫ2∂w

∂y

= −α√

u2+v22w2 1

ǫu+ǫ∂b

∂xw

N(b, ǫ),

∂b

∂x

νh ∂v

∂x+∂u

∂y

+ ∂b

∂y

he)∂v

∂y −(νh−νe)∂u

∂x−2νe∂w

∂z

+

∂b

∂y 2

− 1 ǫ2

! νv

∂v

∂z +ǫ2∂w

∂y

+ ∂b

∂x

∂b

∂y

νv ∂u

∂z +ǫ2∂w

∂x

= −α√

u2+v22w2 1

ǫv+ǫ∂b

∂yw

N(b, ǫ).

(12)

where

N(b, ǫ) = s

1 +ǫ2(∂b

∂x)22(∂b

∂y)2.

(9)

3 Derivation of the setion-averaged shallow water model

3.1 Seond order approximation in ǫ

Inorderto derive oursetion-averaged shallowwater model,a numberofapproximations have

to beperformed. Firstly,we assumethat the vertial eddyvisosityis rst orderwith respet

to theratio between thevertialand longitudinalsales, thatis,

νv = ǫ νv,0, (13)

whereνv,0isagivenpositivequantity. Thisassumptionanbejustiedbyasimpledimensional analysis. Indeed,followingthePrandtlhypothesis,theeddyvisosityishomogeneoustoalength

times aveloity,and more preisely

µ

ρ ∼ l2m||D||, (14)

where lm is the mixing length of the turbulent ow and ||D|| is the norm of the strain-rate tensor. When onsidering the vertial eddy visosity, lm is homogeneous to a depth and the strain-rate tensor reduesto the vertial aeleration, therefore weonlude that

µ ρ ∼ l2m

∂U

∂z

. (15)

NotethatPrandtl'smixinglength modelseefor instaneRef. [11 ℄ isbasedonthis assump-

tion. Adimensionalizing thisexpression ofµv gives:

U Lµˆv ∼ H2m2 s

U2 H2

(∂uˆ

∂z)2+ (∂ˆv

∂z)2+ (ǫ∂wˆ

∂z)2

∼ U H lˆm2

∂uˆ

∂z

, (16)

where the hat denotes herethe adimensional variables. Thus

νv = µˆv

ρU L ∼ ǫ lˆm2 ρ

∂ˆu

∂z

= O(ǫ). (17)

Moreover, the horizontal and bulk visosities are of same order as the vertial eddy visosity,

and thereforewe an write:

νh = ǫ νh,0, νe = ǫ νe,0, (18)

where νh,0 and νe,0 are two given positive quantities. Finally, we assume a slow varying bathymetry in the longitudinal diretion, as it has been done often inthese derivations see

for instaneRef. [5℄ ,and we onsideraonstant atmospheripressure, that is

∂b

∂x =O(ǫ) and ∇pa= 0. (19)

Sine our aim is to obtain a seond order approximation with respet to ǫ of the three-

dimensional system, we neglet quantities of order O(ǫ2). In this way, under the previous

(10)

assumptions, (10) beomes:













































































∂u

∂t +∂u2

∂x +∂uv

∂y +∂uw

∂z + ∂p

∂x = ǫ ∂

∂x

h,0e,0)∂u

∂x−(νh,0−νe,0)∂v

∂y

+ ǫ ∂

∂y

νh,0(∂u

∂y + ∂v

∂x)

+1 ǫ

∂z

νv,0

∂u

∂z

+ǫ ∂

∂z

νv,0

∂w

∂x

,

∂v

∂t +∂uv

∂x + ∂v2

∂y +∂vw

∂z +∂p

∂y = ǫ ∂

∂x

νh,0(∂u

∂y + ∂v

∂x)

+ ǫ ∂

∂y

h,0e,0)∂v

∂y−(νh,0−νe,0)∂u

∂x

+1 ǫ

∂z

νv,0

∂v

∂z

+ ǫ ∂

∂z

νv,0

∂w

∂y

,

∂p

∂z = −G+ǫ ∂

∂x

νv,0∂u

∂z

+ǫ ∂

∂y

νv,0∂v

∂z

+ǫ ∂

∂z

e,0∂w

∂z

,

∂u

∂x + ∂v

∂y + ∂w

∂z = 0.

(20)

together withboundary onditions onthefree surfaeΓs(t),





















































∂η

∂t +u∂η

∂x +v∂η

∂y = w,

∂η

∂x

p−ǫ(νh,0e,0)∂u

∂x +ǫ(νh,0−νe,0)∂v

∂y

−∂η

∂y

ǫ νh,0(∂v

∂x+∂u

∂y)

+1 ǫνv,0

∂u

∂z +ǫ νv,0

∂w

∂x = pa∂η

∂x,

−∂η

∂x

ǫ νh,0

∂v

∂x +∂u

∂y

+∂η

∂y

p−ǫ(νh,0e,0)∂v

∂y +ǫ(νh,0−νe,0)∂u

∂x

+1 ǫνv,0

∂v

∂z +ǫ νv,0

∂w

∂y = pa∂η

∂y,

−∂η

∂x(ǫ νv,0

∂u

∂z)−∂η

∂y(ǫ νv,0

∂v

∂z)−p+ 2ǫνe,0

∂w

∂z = −pa,

(21)

(11)

and nearthe bottom at z=zr,





























































 u∂b

∂x+v∂b

∂y =w,

∂b

∂x

ǫ(νh,0e,0)∂u

∂x−ǫ(νh,0−νe,0)∂v

∂y −2ǫ νe,0

∂w

∂z

+ ∂b

∂y

ǫ νh,0

∂v

∂x+∂u

∂y

− 1 ǫνv,0

∂u

∂z +ǫ νv,0

∂w

∂x

= −α

ǫ ||u||u,

∂b

∂x

ǫνh,0

∂v

∂x +∂u

∂y

+ ∂b

∂y

ǫ(νh,0e,0)∂v

∂y−ǫ(νh,0−νe,0)∂u

∂x

− 2ǫ νe,0

∂w

∂z

+

(∂b

∂y)2− 1

ǫ2 ǫ νv,0

∂v

∂z

= −α||u||

1

ǫv+ǫ∂b

∂yw

,

(22)

where u = (u, v) is thehorizontal veloity. Notiethat, negleting termsinO(ǫ) in (22)2 and

(22)3,we obtain thelassialboundaryondition onthebottom (seee.g. Ref. [2 ℄):

νv,0

∂u

∂z = −α||u||u .

3.2 Vertial integration of the equations

We willnowvertially-integrate system(20)between thefreesurfae andthereferenelevelzr

atwhih thebottomonditions aregiven. We willdenotebyh theorretedwaterdepth,that

isthereal water depthorretedbythe distane∆zr: h(x, y, t) = η(x, y, t)−zr(x, y)

= η(x, y, t)−b(x, y)−∆zr. (23)

Then, for any three-dimensional variable f, we denote withthe average along the vertial

diretion,

f¯(x, y, t) = 1 h(x, y, t)

Z η zr

f(x, y, z, t)dz.

Letus rst vertially-integrate the momentum equation(20)1 for ubetween thereferenebot-

tomlevel zr and thefreesurfae. Making useof the Leibnitzrule yields:

Références

Documents relatifs

From this derivation, we prove that solutions of the water waves equations, associated with well-prepared initial data, can be approximated in the shallow water regime, to within

As the the present paper is focused on the drag coefficient modelling, the choice of an ad-hoc model based on mixing length concept ensures a representative model for the

The complexity of the solution of the function estimation problem using the SV representation depends on the com- plexity of the desired solution (i.e. on the required number of SVs

in Lagrangian coordinates in slab symmetry with a source term and the equations in Eulerian coordinates in plane, cylindrical or spherical symme- try. Explicit formulae

The latter theory car- ries over to the theory of the quasilinear and fully nonlinear operators considered here, and hence from the bounds for the fundamental solution we deduce

Then, the pattern matching is initiated and SDMLib tries to find model objects of type Variable that are connected to the current Activity object via a locals link. If there

Recent space missions like lunar Smart-1 mission, Boeing orbital transfer, using electric propulsion are innovative design feature to reduce launch costs and lead to the analyse of

Note also that the solutions of the Monge problem in many geodesic space X can be disintegrated along transport rays, i.e parts of X isometric to intervals where the obtained