Editorial - Colloque sur l’application industrielle de la thermodynamique moléculaire.

of processing evaluation of risk in production hydrate evaluation of risk in gas and the use of such as CPA and SAFT for of


INTRODUCTION
La meilleure compre´hension des phe´nome`nes a`l'e´chelle mole´culaire ouvre la voie a`un vaste champ de nouvelles applications possibles pour l'industrie chimique et au-dela`. Les efforts re´cents dans ce domaine ont permis de cre´er des mode`les, des me´thodes de simulation et des outils non seulement a`meˆme de re´soudre des proble`mes acade´miques mais aussi de contribuer substantiellement aux projets industriels de recherche et de de´veloppement. Ils ouvrent la voie a`une meilleure compre´hension et a`l'ame´lioration de proce´de´s qui, jusqu'ici, ne pouvaient eˆtre appre´hende´s que de manie`re empirique.
Le besoin de proce´de´s plus efficaces et plus respectueux de l'environnement est un moteur important de l'innovation industrielle. L'utilisation croissante de la compre´hension des phe´nome`nes mole´culaires dans les applications thermodynamiques a transforme´ce domaine en un champ de recherche primordial pour la de´couverte de nouveaux concepts et applications : -les principes thermodynamiques, associe´s a`des concepts de me´canique statistique et a`la puissance de calcul de´ja`facilement disponible, permettent une compre´hension de´taille´e des phe´nome`nes a`l'e´chelle atomique et mole´culaire, en utilisant des mode`les mole´culaires de plus en plus re´alistes. L'utilisation simultane´e de me´thodes aux e´chelles me´so et macro permet la simulation multi-e´chelle de proce´de´s complexes, fournissant ainsi un outil cleṕ our faciliter la conception des proce´de´s ; -ces me´thodes de simulation ont aussi mene´au de´veloppement de nouvelles e´quations d'e´tat, fonde´es sur la physique mole´culaire, qui rendent possible l'introduction d'une puissance pre´dictive croissante dans les simulateurs de proce´de´s et qui ouvrent la voie a`de vraies me´thodologies de conception des produits ; -l'e´tendue des de´veloppements expe´rimentaux est aussi accrue via l'utilisation de donne´es comple´mentaires (densite´, calorime´trie, spectroscopie, etc.) afin d'assurer une repre´sentation comple`te et cohe´rente de la structure microscopique des fluides et du comportement macroscopique de phase.

LE COLLOQUE INMOTHER
InMoTher (Application industrielle de la thermodynamique mole´culaire) a e´te´mis en place par le groupe de travail de l'EFCE sur la Thermodynamique et les Proprie´te´s de Transport. Il s'est tenu a`l'E´cole Normale Supe´rieure (ENS) de Lyon le 19 et le 20 mars 2012 et a e´teó rganise´de manie`re conjointe par la SFGP (Socie´te´Franc¸aise de Ge´nie des Proce´de´s), l'EFCE (Fe´de´ration Europe´enne du Ge´nie Chimique), le groupe de travail allemand ProcessNet sur la Thermodynamique (repre´sente´par Dechema dans l'organisation) et l'ENS. Son but e´tait de fournir aux experts industriels une vision claire des nouvelles opportunite´s lie´es au de´veloppement rapide des travaux interdisciplinaires, alliant les efforts des sciences naturelles et de l'inge´nierie. Au total 159 participants venant de 22 pays diffe´rents ont pu participer a`ce colloque. 58 de ces participants e´taient issus du monde de l'industrie.
Le soutien de l'industrie fut clairement de´montre´a`travers a`un important parrainage : deux sponsors Or : TOTAL et la re´gion Rhoˆne-Alpes a`travers un important poˆle de compe´titiviteÁ xelera, et 5 sponsors Argent: le CNRS, Air Liquide, Linde Engineering, Rhodia du groupe Solvay et IFP Energies nouvelles, centre de recherche dans lequel le pre´sident de la confe´rence, Jean-Charles de Hemptinne, occupe la chaire de la fondation Tuck « Thermodynamique pour les carburants issus de la biomasse ».

Simulation moléculaire et outils Ab Initio
La premie`re confe´rence ple´nie`re, pre´sente´e par le professeur A. Panagiotopoulos (Universited e Princeton), e´tait intitule´e : « Simulation mole´culaire d'e´quilibres de phase et d'assemblage de phase : progre`s et de´fis ». Cet expose´proposait un bilan des me´thodes de simulation mole´culaire pour la mode´lisation d'e´quilibres et d'auto-assemblage de phase. On peut le re´sumer comme suit : l'ensemble de Gibbs est une approche directe propose´e il y a plus de vingt ans et qui est adapte´e aux calculs de moyenne pre´cision. La me´thode dite « Grand Equilibrium » est base´e sur des simulations isothermes-isobares dans la phase liquide et par l'ensemble pseudo grand canonique dans la vapeur. Combine´es avec l'inte´gration de Gibbs-Duhem introduite par Kofke, ces me´thodes permettent le calcul en quasi-routine des diagrammes de phase pour des potentiels intermole´culaires donne´s. Des me´thodes alternatives, en particulier la me´thode de Monte-Carlo par reponde´ration d'histogrammes, fournissent une bonne pre´cision pre`s des points critiques et une solution pour surmonter les hyste´re´sis et longues e´chelles de temps inhe´rentes a`l'auto-assemblage d'agents tensio-actifs et de polyme`res. Durant ces dernie`res anne´es, avec l'arrive´e de logiciels puissants, extensibles et open-source de calcul de dynamique mole´culaire, les e´chelles de temps sur lesquelles l'auto-assemblage et la se´paration de microphases peuvent eˆtre e´tudie´es, ont e´te´e´tendues jusqu'a`la ls, meˆme dans le cas de mode`les de potentiel re´alistes dans un solvant explicite. Cependant, il existe toujours un besoin important de de´veloppement de mode`les a`mailles grossie`res, ou « coarse-grained », capables de capturer et de structurer la thermodynamique afin d'e´tendre les e´chelles de temps et d'espace des syste`mes pouvant eˆtre simule´s.
La premie`re confe´rence invite´e a e´te´pre´sente´e de manie`re conjointe par S. Lustig de Du Pont et A. Klamt de CosmoLogic. Elle fut de´die´e a`« L'application de COSMO-RS dans la conception de syste`mes ioniques ». On peut la re´sumer ainsi : il est aujourd'hui reconnu que COSMO-RS est une the´orie largement applicable, utilise´e pour pre´dire de manie`re pre´cise une vaste gamme de proprie´te´s des fluides complexes. En utilisant les premiers principes de la chimie quantique, un Hamiltonien de contact empirique et la thermodynamique statistique, COSMO-RS pre´dit les proprie´te´s a`l'e´quilibre de substances pures : pression de vapeur, tempe´rature d'e´bullition, enthalpie de vaporisation, mais aussi de me´langes : pression partielle de vapeur, activite´, solubilite´de gaz, solubilite´de liquides, diagrammes de phase de fluides, pKa, etc. Bien que les me´thodes de dynamique mole´culaire base´es sur un champ de force classique et les me´thodes Monte-Carlo puissent eˆtre utilise´es pour pre´dire ces proprie´te´s, les imple´mentations de COSMO-RS sont bien plus rapides et pre´cises. Les intervenants ont explique´que l'approche the´orique fondamentale avait e´te´de´veloppe´e au de´part pour les syste`mes mole´culaires neutres, et que des travaux plus re´cents ont montreĺ e succe`s de l'utilisation de cette me´thode pour des liquides ioniques complexes. La pre´sentation visait a`re´sumer les fondamentaux de la the´orie COSMO-RS et a`explorer ses applications pour la pre´diction de proprie´te´s et la conception de syste`mes liquides ioniques. Une question centrale est de savoir si les pre´dictions de proprie´te´s de COSMO-RS sont plus pre´cises lorsque les ions mole´culaires sont traite´s de manie`re quantique, en ions se´pare´s, en ions apparie´s ou bien dans les deux cas. Des e´valuations des coefficients d'activite´a`dilution infinie et des coefficients de Loi de Henry a`concentration finie ont e´te´faites. Deux cas d'application ont e´te´conside´re´s : la conception d'un liquide ionique optimal pour les proce´de´s de refroidissement par absorption et la conception de solvants pour les batteries lithium-ion. Dans les deux cas, l'application COSMOtherm e´tait imple´mente´e comme une sous-routine d'un algorithme ge´ne´ral optimisant le crite`re thermodynamique de conception. Comme application pratique, l'intervenant a pre´sente´comment se´lectionner des paires cation-anion dans une e´tude de refroidissement par absorption afin d'optimiser le coefficient thermodynamique de performance tout en minimisant les chimies de de´gradation thermique et d'hydrolyse. Pour l'e´tude de la batterie lithium-ion, la pre´diction de la solubilite´du LiPF 6 et les profils de spe´ciation ionique dans deux classes de solvants organiques tre`s diffe´rentes ont e´te´conside´re´s. Ici, le crite`re thermodynamique est la minimisation de l'e´nergie libre de Gibbs du syste`me, soumise aux contraintes d'e´quilibre d'action de masse, d'e´quilibre des charges et d'e´quilibre solide liquide. Il a e´te´alors montre´que les deux applications demandent des pre´dictions pre´cises des proprie´te´s thermodynamiques pour une gamme de tempe´ratures et de concentrations. Il a e´te´conclu que, dans les deux cas, les simples re´sultats qualificatifs fournissaient une compre´hension physique des phe´nome`nes qui pouvait permettre a`un inge´nieur chimiste de comprendre ces syste`mes complexes et de concevoir des syste`mes utiles.
La seconde pre´sentation industrielle invite´e e´tait consacre´e a`« La thermochimie des mate´riaux industriels pour l'industrie ». Elle fut pre´sente´e par P. Raybaud d'IFP Energies nouvelles. Il a explique´que le contexte environnemental pousse la communaute´de la chimie a`proposer des approches innovantes pour ame´liorer la pre´diction des proprie´te´s des mate´riaux utilise´s dans des applications pour de´velopper des e´nergies nouvelles, par exemple pour produire des carburants plus propres et renouvelables. Pour cela, des concepts rationnels et quantifie´s concernant les proprie´te´s de surface ou de volume e´taient ne´cessaires pour ame´liorer les mate´riaux existant ou en de´couvrir de nouveaux. Cette pre´sentation illustrait comment la mode´lisation mole´culaire ab initio utilisant la the´orie de la fonctionnelle de densite´(DFT), a permis d'approfondir la compre´hension et la pre´diction des proprie´te´s de trois importantes classes de mate´riaux dans le domaine des e´nergies nouvelles : -hydrures solides pour le stockage de l'hydroge`ne, -mate´riaux photocatalytiques a`base de TiO 2 , -surfaces me´talliques catalytiques.
La manie`re dont les diagrammes thermodynamiques de phase e´taient de´termine´s par des calculs ab initio en fonction des conditions d'utilisation (T, P) exprime´es par le potentiel chimique du re´actif/adsorbat a e´te´souligne´e . Concernant les proprie´te´s de volume du mate´riau,  les hydrures solides ont e´te´pre´sente´s comme un premier exemple concret. L'e´valuation DFT  des stabilite´s thermodynamiques et des enthalpies d'hydroge´nation a re´ve´le´les limites et le  potentiel respectif des alanates et des hydrures KSi. Puis en se penchant sur les mate´riaux de  TiO 2 dope´s a`l'azote, il a e´te´annonce´qu'une re´cente e´tude DFT expliquait l'augmentation de l'absorption de la lumie`re visible en fonction du potentiel chimique d'hydroge`ne. Pour terminer, la thermochimie de surface d'un catalyseur me´tallique en pre´sence de pression d'hydroge`ne a e´te´illustre´e. Il a e´te´montre´que ces re´sultats the´oriques fournissent des guides rationnels pour les nouvelles expe´riences. Au-dela`de la thermodynamique mole´culaire, une bre`ve introduction des de´fis de la cine´tique mole´culaire a aussi e´te´donne´e.
La premie`re pre´sentation provenant de l'industrie dans ce domaine a e´te´donne´e par G. Folas de Statoil, dans une confe´rence intitule´e : « Mode`les avance´s dans la pratique industrielle : de la conception a`l'optimisation de proce´de´s ». Il a commence´en expliquant que Statoil, comme beaucoup d'autres entreprises, utilise principalement les simulateurs du commerce pour la conception, l'optimisation et la re´solution des proble`mes des sites de production. Les phases initiales du projet peuvent eˆtre effectue´es de manie`re interne, et, a`partir du moment ou`le concept est arreˆte´, il est e´tudie´de manie`re plus de´taille´e par des socie´te´s d'inge´nierie. Le principe ge´ne´ral est d'utiliser autant que possible des outils standards issus du commerce et facilement accessibles. Afin d'ame´liorer les capacite´s de la compagnie et de passer outre les limitations des simulateurs du commerce, des composants CAPE-OPEN ou des outils inde´pendants (de´veloppe´s de manie`re interne ou en collaboration avec des universite´s) sont utilise´s pour des applications spe´cifiques. Sa pre´sentation exposait des exemples d'application des mode`les avance´s pour concevoir et optimiser des proce´de´s. Toute une gamme d'applications a e´te´discute´e, comme par exemple le niveau de confiance des calculs de la teneur en eau du gaz naturel dans la conception des sites de production, l'e´valuation des risques de corrosion dans les pipelines de production, l'inhibition des hydrates de gaz ou l'e´valuation du risque de gel dans les proce´de´s gazeux a`basse tempe´rature, ainsi que l'utilisation des mode`les tels que CPA et SAFT.
La pre´sentation suivante a e´te´donne´e par M. Heiling de BASF, qui pre´sentait « Les applications industrielles de la thermodynamique mole´culaire : se´lection et extrapolation ». Sa pre´sentation expliquait que les donne´es des proprie´te´s physiques et la thermodynamique sont a`la base du de´veloppement des proce´de´s et des applications d'inge´nierie chimique. Le point de de´part de tout proce´de´est et restera, dans un futur proche, la disponibilite´de donne´es expe´rimentales. Une base de donne´es comple`te des proprie´te´s physiques est donc essentielle. Ces donne´es sont utilise´es comme base pour ajuster les e´quations, permettant un acce`s rapide aux donne´es pertinentes du proce´de´mais aussi une manie`re de construire ide´es et concepts. Deux aspects ont e´te´souligne´s : la se´lection et l'extrapolation des donne´es de proprie´te´s physiques. Les me´thodes thermodynamiques peuvent eˆtre applique´es a`la se´lection de proce´de´s et de solvants fonctionnels. Afin de convertir un concept de proce´de´ou une application fonctionnelle en des crite`res de proprie´te´s physiques, les donne´es disponibles dans les bases de donne´es doivent eˆtre comple´te´es par une mode´lisation pre´dictive de multiples composants. Il a e´te´montreq ue, en plus des me´thodes de contribution de groupe, le mode`le COSMO-RS de re´solution du continuum chimique quantique peut eˆtre tre`s utile. L'extrapolation signifie la pre´diction de donne´es dans de larges gammes de tempe´rature, de pression et de concentration ainsi que la pre´diction de syste`mes complexes, base´e sur les donne´es des composants purs et binaires. M. Heilig a montre´que les e´quations d'e´tat de type CPA et PCSAFT permettent une extrapolation physiquement plus correcte. Cependant une parame´trisation complexe des composants purs est ne´cessaire. Une simulation mole´culaire, base´e sur les potentiels intermole´culaires obtenus a`partir de moins de donne´es thermodynamiques, requiert du temps et des efforts et n'est pour l'instant conside´re´e que pour des syste`mes importants ou`aucune autre solution n'existe. En plus d'un ensemble quasiment complet de donne´es de proprie´te´s physiques, la simulation mole´culaire permet d'acce´der aux informations concernant la structure des fluides ou des me´langes de fluides.

Outils moléculaires pour la gestion des données
La troisie`me confe´rence ple´nie`re, dont le titre e´tait : « Moteur ThermoData du NIST : Augmentant la valeur, diminuant la « pollution », e´largissant le cadre et fournissant un moyen de communication pour les informations de proprie´te´s thermodynamiques » e´tait donne´e par M. Frenkel du NIST. Celui-ci a affirme´que le moteur ThermoData du NIST (ThermoData Engine, TDE) repre´sentait la premie`re imple´mentation a`grande e´chelle du concept d'e´valuation dynamique de donne´es pour les donne´es de proprie´te´s thermophysiques et thermochimiques. Il a explique´que ce concept demandait le de´veloppement de vastes bases de donne´es e´lectroniques, capables de stocker toutes les donne´es pertinentes connues a`ce jour avec les descriptions de´taille´es des me´tadonne´es et des incertitudes. La combinaison de ces bases de donne´es e´lectroniques avec un logiciel intelligent (syste`me-expert), conc¸u pour se´lectionner automatiquement des donne´es expe´rimentales et pre´dites disponibles, permet d'offrir la capacite´de produire des donne´es e´value´es de manie`re critique et automatique ou « sur commande ». Le domaine d'application de TDE inclut les compose´s purs, les me´langes binaires, les me´langes ternaires et les re´actions chimiques. TDE est un composant critique du Syste`me Mondial d'Information en Sciences et Inge´nierie (Global Information System in Science and Engineering, GISSE). Le roˆle et la faisabilite´de l'utilisation d'une grande varie´te´d'outils mole´culaires (tels que l'identificateur de syste`me chimique InChI) et de me´thodes de pre´diction des proprie´te´s (contribution de groupe, QSPR avec technologie SVM, Monte-Carlo et ab initio) dans l'e´valuation de donne´es critiques, ainsi que dans le processus global de validation des donne´es, incluant des journaux scientifiques majeurs dans le domaine et supporte´s par TDE, ont e´te´discute´s. Diverses options pour inclure TDE dans des logiciels d'inge´nierie, y compris dans les moteurs de conception de proce´de´s chimiques, ont e´te´illustre´es. Dans sa conclusion, M. Frenkel a explique´que la technologie TDE a e´teÉ´d itorial incorpore´e a`un logiciel en ligne afin d'aider le processus de planification des expe´riences de mesure des proprie´te´s. Le logiciel en acce`s gratuit et libre sur internet est conc¸u pour eˆtre utilise´par des expe´rimentateurs, partout dans le monde.

Des outils corrects pour une utilisation correcte
Toutes les parties e´taient d'accord pour dire que l'utilisation d'outils thermodynamiques requie`re une expertise spe´cifique. Alors que certaines entreprises ont engage´un spe´cialiste dont la responsabilite´principale est de de´velopper des applications en interne, d'autres conside`rent que le niveau d'expertise requis est trop e´leve´compte tenu des be´ne´fices possibles, tout du moins pour le moment. Diffe´rents objectifs peuvent eˆtre envisage´s : -le de´veloppement d'e´quations d'e´tat pre´dictives peut profiter de la possibilite´offerte par la simulation nume´rique d'isoler des phe´nome`nes spe´cifiques, affinant ainsi l'e´quation pour un besoin spe´cifique. Cette approche permet d'ame´liorer syste´matiquement et rationnellement la repre´sentation du syste`me et de transfe´rer les caracte´ristiques mole´culaires a`un mode`le plus grossier (coarse-grained) comme l'e´quation d'e´tat ; -l'utilisation de l'e´quation d'e´tat pose souvent une difficulte´spe´cifique : le parame´trage ade´quat. Ici des mode`les plus fondamentaux peuvent eˆtre utiles, soit pour ge´ne´rer des donne´es pseudo-expe´rimentales afin de comple´ter les bases de donne´es, soit pour aider a`la de´termination des parame`tres a`partir de leur signification atomistique (utilisation de me´thodes ab initio) ; -les approches nume´riques telles que COSMO-RS sont principalement utilise´es dans un but de se´lection. Puisque ces me´thode sont fonde´es sur des calculs ab initio et ne requie`rent pas de parame`tres empiriques, leur utilisation permet d'explorer rapidement de nombreuses situations pour lesquelles peu ou pas de donne´es existent ; -les me´thodes a`grande e´chelle, qui sont des approches gros grain (coarse-grained) de simulations mole´culaires, repre´sentent un autre groupe de me´thodes qui commence a`devenir tre`s utile, pour explorer des e´chelles de temps et de distance adapte´es aux applications technologiques. Ces e´chelles peuvent ne pas encore eˆtre explore´es via des me´thodes de simulation atomistique.

Cycle de vie du procédé
Les grandes entreprises, qui de´veloppent de nouveaux proce´de´s, insistent sur l'importance d'utiliser diffe´rentes approches qui de´pendent de l'e´tape du cycle de vie du de´veloppement du proce´de´.
-dans les phases initiales, un outil de se´lection est important. Ici, la pre´cision n'est pas critique et des logiciels autonomes peuvent eˆtre utilise´s. C'est ici que les logiciels de type simulation mole´culaire sont adapte´s ; -dans un second temps, lorsque la conception du proce´de´commence, la ge´ne´ration de don-ne´es est importante. Dans ce cas, la simulation mole´culaire est utilise´e en mode « production » et peut eˆtre utilise´e pour comple´ter les expe´riences afin de re´duire le temps ne´cessaire pour produire les donne´es. Par ailleurs, un outil pouvant eˆtre interface´avec un simulateur de proce´de´est alors essentiel. La pre´cision n'est pas critique et il est plus important d'observer des tendances correctes. Les mode`les de contribution de groupes trouvent ici leur place naturelle. CAPE-OPEN n'a pas e´te´mentionne´durant les discussions, mais pourrait eˆtre tre`s utile a`ce niveau ; -dans une e´tape finale, la pre´cision est critique et un mode`le spe´cifique doit eˆtre utilise´. La plupart du temps, a`cause de longues habitudes et de la confiance porte´es aux anciens mode`les a`corre´lations, il est tre`s difficile de convaincre les inge´nieurs proce´de´s d'utiliser de nouveaux mode`les : il faut pour cela qu'ils apprennent a`leur faire confiance. Le crite`re pour cela peut beaucoup de´pendre du type d'industrie conside´re´e (pe´trochimique, chimique, pharmaceutique, etc.).

Collecte des données et compréhension des phénomènes
Les participants acade´miques insistent sur le fait que leur effort est plus centre´sur la compre´hension des phe´nome`nes sous-jacents que sur la production de donne´es. Ils reconnaissent que la collecte des donne´es est un travail important, mais affirment qu'il est difficile de publier ces re´sultats. Il est surprenant qu'aujourd'hui, les donne´es soient parfois pre´sen-te´es sans re´fe´rencer les auteurs originaux mais plutoˆt (dans le meilleur des cas !) la base de donne´es dont elles sont issues. Ce me´pris des normes scientifiques ne peut pas eˆtre tole´re´par la communaute´scientifique. Ceci est particulie`rement faˆcheux pour les groupes qui publient des donne´es expe´rimentales qui demeurent essentielles pour re´gler les parame`tres des mode`les. C'est pourquoi de saines comple´mentarite´s doivent eˆtre de´veloppe´es entre les diffe´rents domaines.
Par conse´quent, il est important de rester a`la pointe des connaissances, afin que l'outil le plus pertinent puisse eˆtre recommande´selon les besoins. C'est un roˆle essentiel que doivent jouer les vendeurs de logiciel. Il n'est souvent pas facile d'apporter une re´ponse, car elle demande une vision globale et des connaissances partant des fondamentaux et allant jusqu'aux applications. Ce proble`me est lie´a`la discussion pre´sente´e pre´ce´demment concernant le cycle de vie des proce´de´s : des approches diffe´rentes seront recommande´es en fonction de l'e´tape de de´veloppement du proce´de´.

Paramétrage
Cette discussion concernant les mode`les peut aussi eˆtre analyse´e du point de vue du para-me´trage : doit-on de´velopper des parame`tres pour des applications spe´cifiques (plus pre´cis) ou pour un domaine d'applications large ? La re´ponse n'est pas unique mais nous pousse ar e´fle´chir sur la pre´cision acceptable, qui peut eˆtre tre`s variable. Ceci implique : -qu'il soit possible d'e´valuer la pre´cision du mode`le, ce qui nous rame`ne a`la disponibilited es donne´es et a`leur pre´cision (qui est rarement disponible !) ; -que les effets des incertitudes du mode`le thermodynamique sur la simulation finale de proce´de´soient connus : c'est rarement le cas.
Il existe de nombreux exemples industriels pour lesquels les re´sultats de simulations de proce´de´s sont sensibles au parame´trage spe´cifique utilise´dans les mode`les. Les questions qui se posent en vue de choisir les donne´es qui serviront au parame´trage sont : quelles sont les donne´es de bonne qualite´, quelles sont celles moins fiables ? Comment doit eˆtre effectueĺ 'ajustement des parame`tres ? Ainsi, la question de l'e´valuation des donne´es des proprie´te´s thermophysiques doit jouer un roˆle de plus en plus important dans la thermodynamique industrielle.
Les simulations mole´culaires ont deux inte´reˆts : aider a`la compre´hension des phe´nome`nes et ge´ne´rer des donne´es pseudo-expe´rimentales. Concernant le premier point, des re´sultats qualitatifs sont attendus et la simulation doit permettre d'identifier des tendances et d'ame´liorer les connaissances concernant de nouveaux compose´s ou de nouveaux domaines de recherche (biomasse, compose´s oxyge´ne´s, fluore´s, etc.). Elle est alors un outil ide´al pour de´velopper des outils a`l'e´chelle supe´rieure : il est essentiel d'e´tudier de nouvelles voies a`meˆme de connecter les diffe´rentes e´chelles de temps et d'espace. Dans la pratique, des outils ab initio de parame´trage des mode`les de grande e´chelle (gros grain) peuvent eˆtre utilise´s ; dans le sens oppose´, il peut eˆtre e´galement possible d'utiliser un mode`le a`grand grain pour initialiser un mode`le a`grain plus petit.
Un e´quilibre doit eˆtre trouve´entre l'utilisation optimale des nouvelles architectures informatiques, ce que peut ne´cessiter une re´e´criture de code, et le besoin de capitaliser sur les anciens programmes qui re´pondent bien au besoin. La re´ponse peut eˆtre assez diffe´rente suivant la complexite´des algorithmes (la re´solution des e´quations d'e´tat peut demander des algorithmes plus complexes que les calculs ab initio ou de simulation mole´culaire).
Si la re´e´criture d'une portion du code est ne´cessaire, il est important de se poser la question du langage de programmation : la plupart des codes disponibles aujourd'hui sont e´crits en Fortran, alors que les nouveaux langages oriente´s-objet peuvent eˆtre plus faciles aè ntretenir et a`interfacer. Le choix syste´matique de la plateforme Windows dans l'industrie peut eˆtre un facteur limitant. Le choix final doit eˆtre un compromis entre efficacite´du temps de calcul, temps de de´veloppement, maintenance et re´sultats attendus.
Il est bon par contre de garder a`l'esprit que les avance´es les plus cruciales sont lie´es au facteur humain : c'est-a`-dire la fac¸on dont l'utilisateur interpre`te les re´sultats. Le bon sens peut parfois bien plus aider a`la compre´hension que de nombreux calculs, bien qu'il faille rester vigilant face aux fausses « bonnes intuitions » et ide´es pre´conc¸ues, qui peuvent eˆtre errone´es parce qu'e´tablies a`une e´chelle diffe´rente.
Il a e´te´montre´que l'acceptation de nouveaux outils demande la confiance de l'utilisateur. C'est possible a`travers une formation spe´cifique.

Modèles conventionnels
Les mode`les conventionnels ou standards utilisent les approches classiques de la thermodynamique applique´e comme celles lie´es aux e´quations d'e´tat cubiques, aux mode`les de coefficient d'activite´ou aux approches de contribution de groupes. Ces me´thodes sont souvent conside´re´es comme matures, mais sont constamment ame´liore´es et/ou applique´es de nouvelles manie`res.
A. Zaitseva et V. Alopaeus ont propose´une nouvelle approche pour ame´liorer la pre´cision des estimations des me´thodes de contribution des groupes (pour les proprie´te´s de compose´s purs). Leur approche, qui exploite la similarite´chimique des compose´s dont les proprie´te´s sont estime´es, est base´e sur ce que l'on appelle une optimisation ponde´re´e par la distance, par opposition a`l'optimisation non-ponde´re´e souvent utilise´e. La me´thodologie propose´e est ge´ne´rique et peut eˆtre applique´e aux diffe´rentes me´thodes de contribution de groupes. Les auteurs ont applique´leur approche avec succe`s aux me´thodes bien connues de Joback-Reid et de Marrero-Gani pour la de´termination du point normal d'e´bullition, ainsi que pour estimer plusieurs autres proprie´te´s dans le cas de la me´thode de Joback-Reid. Dans tous les cas, les estimations sont ame´liore´es par rapport aux me´thodes originales. Cette nouvelle approche pourrait s'ave´rer encore plus utile pour les me´thodes de contribution de groupes utilisant des groupes de second et troisie`me ordres. . Les auteurs ont conside´re´a`la fois une version du mode`le utilisant des parame`tres spe´cifiques aux sels et une autre avec des parame`tres spe´cifiques aux ions, cette dernie`re pre´sentant de manie`re ge´ne´rale de meilleurs re´sultats. Le mode`le repre´sente de manie`re satisfaisante, avec peu de parame`tres ajustables, les pressions de vapeur, les coefficients osmotiques et coefficients d'activite´ionique moyens de 13 solutions salines aqueuses 1:1 sur une gamme de tempe´rature allant jusqu'a`363,15 K. De bonnes pre´dictions pour les pressions de vapeur de me´langes de deux sels ont aussi e´te´pre´sente´es.

Simulation moléculaire et COSMO-RS
Les simulations mole´culaires et les approches nume´riques de chimie quantique sont devenues tre`s populaires durant les 20 dernie`res anne´es et ont e´te´applique´es a`divers syste`mes.

CONCLUSIONS
Ce colloque a e´te´une excellente occasion d'e´changes entre les experts de la thermodynamique mole´culaire issus de diffe´rentes origines. En tant qu'activite´sponsorise´e par le groupe de travail « Thermodynamique et proprie´te´s de transport » de l'EFCE, il peut eˆtre conside´reć omme une e´tape positive vers la cre´ation d'une communaute´dans ce domaine, avec une participation active de l'industrie.
Diffe´rents points pouvant ne´cessiter du travail dans l'avenir ont e´te´souleve´s : -de nombreuses me´thodes existent, et leur pertinence de´pend e´norme´ment de leur utilisation finale. Les diffe´rents outils ne sont clairement pas en compe´tition, mais doivent plu-toˆt eˆtre vus comme comple´mentaires. Il pourrait eˆtre utile de proposer une vue d'ensemble des derniers de´veloppements et des meilleurs manie`res d'employer les comple´mentarite´s entre les outils ; -tre`s peu d'efforts sont investis aujourd'hui dans l'e´tude des liens complexes qui existent entre la qualite´des donne´es expe´rimentales utilise´es pour le parame´trage et les incertitudes qui en re´sultent dans les re´sultats finaux. Ceci ame`ne a`une certaine de´valuation de travaux expe´rimentaux de haute qualite´, dont la valeur doit eˆtre reconnue. La question de l'utilisation des outils de simulation mole´culaire pour produire des pseudodonne´es, et par conse´quent l'introduction de ces donne´es dans les bases de donne´es commerciales usuelles se pose aussi ; -il est ne´cessaire de de´velopper plus avant les me´thodologies d'augmentation de l'e´chelle des approches nume´riques : comment naviguer entre les diffe´rentes e´chelles et mode`les thermodynamiques, de quantique a`mole´culaire, me´soscopique ou de volume.

REMERCIEMENTS
De nombreux participants ont contribue´de manie`re directe a`ce re´sume´, soit en prenant des notes durant la re´union ou en fournissant des commentaires. Les auteurs souhaitent remercier particulie`rement les contributions de C. Coquelet

INTRODUCTION
The improved understanding of molecular-scale phenomena opens up an immense field of new possible applications for the chemical industry and beyond. The recent efforts in this field have resulted in models, simulation methods, and tools that allow not only solving academic problems but also contribute substantially to industrial research and development. They open the route to gaining insight and improving processes that up to now could only be dealt with empirically.
A large drive for industrial innovation is created by the need for more efficient and ecofriendly processes. The increased use of molecular understanding in thermodynamic applications makes that this science becomes a primary field for new concepts and applications: -the thermodynamic principles coupled with statistical mechanical concepts and a readily available computer power allows a detailed understanding of atomic and molecular scale phenomena, using increasingly realistic molecular models. The simultaneous use of mesoscale and macro-scale methods opens multiscale simulation of complex processes yielding a key tool to assist process design; -these simulation methods have also led to the development of new molecular based equations of state that make it possible to introduce increased predictive power into the process simulators, and open the way for true product design methodologies; -the scope of the experimental developments is also enhanced through the increased use of complementary data (density, calorimetry, spectroscopy, etc.) so as to ensure a coherent and complete picture of the microscopic fluid structure and macroscopic phase behavior.
Several reviews of the industrial perspectives with respect to these developments have been published. A recent review [1] was initiated by the working party "Thermodynamics and Transport Properties" of the European Federation of Chemical Engineers (EFCE). One of the conclusions was the need to create exchange opportunities between the various professionals that work in this field. The workshop entitled "Industrial Use of Molecular Thermodynamics" (InMoTher), is an outcome of this reflexion. This special issue of OGST is an opportunity to publish a number of papers that have been presented, and in this editorial, we want also to report the main conclusions of the discussions that have taken place, in the same way as was done for a similar workshop on experimental thermodynamics (D. Richon 60th birthday and departure workshop) [2]. All the presentations at the InMoTher workshop are also accessible on the SFGP web site, at the internet address: http://www.sfgp.asso.fr/?cat=menu&mcat=group&id=130.

THE INMOTHER WORKSHOP
The InMoTher workshop (Industrial Use of Molecular Thermodynamics) was initiated by the EFCE working party on Thermodynamics and Transport Properties. It was held at the Editorial "E´cole Normale Supe´rieure" (ENS) in Lyon on 19 and 20 March, 2012, jointly organized by the SFGP (Socie´te´Franc¸aise de Ge´nie des Proce´de´s), the EFCE (European Federation of Chemical Engineers), the German ProcessNet working party on Thermodynamics (represented by Dechema in the organization) and the ENS. It aimed at providing to the industrial experts a clear vision into the opportunities that lie within this rapidly developing interdisciplinary field, in which efforts from natural sciences and engineering are combined. A total of 159 participants have registered from 22 different countries. 58 of these participants originated from industry.
The support of industry was clearly expressed through a strong sponsorship: two gold sponsors, TOTAL and the Rhoˆne-Alpes region through the Axelera competitiveness cluster, and five silver sponsors, which are CNRS, Air Liquide, Linde Engineering, Rhodia from the Solvay group and IFP Energies Nouvelles from where the conference chairman, Jean-Charles de Hemptinne, holds the Tuck foundation chair for Biofuels Thermodynamics.

The Plenary and Parallel Sessions
Three plenary review sessions were organized around the three major competences in industrial thermodynamics: molecular simulation, molecular equations of state and molecular tools for data management. Six industrial invited speakers have provided their vision in the parallel sessions.

Molecular Simulation and Ab Initio Tools
The first plenary lecture, given by Prof. A. Panagiotopoulos (Princeton University), was entitled: "Molecular simulation of phase equilibria and phase assembly: progress and challenges". This lecture provided a review of molecular simulation methods for modeling phase equilibria and self-assembly and may be summarized as follows: the Gibbs ensemble is a direct approach proposed over twenty years ago that is suitable for calculations of moderate accuracy. The grand equilibrium method is based on isothermal-isobaric and pseudo grand canonical simulations in the liquid and vapor phases, respectively. Combined with Kofke's Gibbs-Duhem integration, these methods allow the relatively routine calculation of phase diagrams for given intermolecular potentials. Alternative methods, in particular histogram reweighting Monte-Carlo provide good accuracy near critical points and provide paths for overcoming hysteresis and long time scales inherent in self-assembly of surfactants and polymers. In recent years, with the advent of fast, scalable, open-source packages for molecular dynamics calculations the time scales over which self-assembly and microphase separation can be studied have been extended to ls, even for realistic potential models with explicit solvent. However, there is still a significant need for development of coarse-grained models that correctly capture and structure thermodynamics in order to extend the time and length scales of systems that can be simulated.
The first invited lecture was jointly presented by S. Lustig from Du Pont and A. Klamt from CosmoLogic. It was devoted to the "Application of COSMO-RS in the Design of Ionic Systems". The lecture can be summarized as follows: it is today acknowledged that COSMO-RS is a broadly applicable theory used for predicting accurately a wide range of properties in complex liquid mixtures. Applying first principles quantum chemistry, an empirical contact Hamiltonian and statistical thermodynamics, COSMO-RS predicts equilibrium pure substance properties: vapour pressure, boiling temperature, enthalpy of vaporization; and mixture properties: partial vapour pressures, activities, gas solubilities, liquid solubilities, fluid phase diagrams, pKa, and more. While classical forcefield based molecular dynamics and Monte-Carlo methods can be used to predict these properties, implementations of COSMO-RS is much faster and more accurate. The speakers explained that the basic theoretical approach was originally developed for neutral molecular systems and that more recent works illustrated successful application with complex ionic liquids. The lecture aimed at summarizing the fundamental underpinnings of COSMO-RS theory and at exploring its application to both property prediction benchmarks as well as design of practical ionic liquid systems. A central question for benchmark studies is whether COSMO-RS property predictions are more accurate when molecular ions are treated quantum mechanically as either separate ions or paired ions or both. Benchmarks involving infinite dilution activity coefficients and finite-concentration Henry's Law coefficients were compared. Two application studies were considered: the design of an optimal ionic liquid for absorption cooling processes and design of solvents for lithium ion batteries. In both cases the COSMOtherm application was implemented as a subroutine within an overall algorithm that optimizes a thermodynamic design criterion. As a practical application, it was considered how to select cation-anion pairs in an absorption cooling study to optimize the thermodynamic coefficient of performance while minimizing thermal and hydrolysis degradation chemistries. For the lithium ion battery study the prediction of temperature-dependent solubility of LiPF 6 and ionic speciation profiles in two very different classes of organic solvents was considered. Here the thermodynamic criterion is the minimization of the system Gibbs free energy subject to mass action balance, charge balance and solid liquid equilibrium constraints. It was shown that both applications required accurate prediction of thermodynamic properties over a range of temperature and finite concentrations. It was concluded that -in both cases -even the qualitative results provided physical understanding that enabled a chemical engineer to understand these complex systems and design useful systems.
The second industrial invited lecture in this topic was devoted to "ab initio thermochemistry of industrial materials for energy" and presented by P. Raybaud from IFP Energies nouvelles. He explained that the environmental context prompts the chemical community to propose innovative approaches for improving the prediction of properties of materials used in new energy applications such as cleaner and renewable fuels. For that purpose, rational and quantified concepts on bulk and surface properties of materials are needed for the improvement of existing materials or the discovery of new ones. This lecture illustrated how ab initio molecular modeling within Density Functional Theory (DFT), has brought insights into the understanding and property prediction of three important classes of materials in the field of new energies: -hydride solids for hydrogen storage, -photocatalytic TiO 2 materials, -metallic catalyst surfaces.
It was highlighted how the thermodynamic phase diagrams are determined by ab initio calculations as a function of the (T, P) working conditions expressed by the chemical potential of the reactant/adsorbate. Considering the properties of bulk materials, hydride solids represent a first relevant example. The DFT evaluation of their thermodynamic stabilities and enthalpies of hydrogenation has revealed the limits and potentialities of alanates and KSi-hydride, respectively. Then, regarding nitrogen-doped TiO 2 materials, it was stated that a recent DFT study explained the origin of their enhanced visible light absorption as a function of the chemical potential of nitrogen. Finally, the surface thermochemistry of a metallic catalyst in the presence of hydrogen pressure was illustrated. Moreover, it was shown that these theoretical results provided rational guides for new experiments. Beyond molecular thermodynamics, a brief opening on molecular kinetic challenges was also given.

Molecular Equations of State
The second plenary lecture, entitled: "SAFT equations of state for complex fluids: model development and applications" was given by J. Gross (Stuttgart University) who explained Editorial that thermodynamicists are faced with the mandate to provide methods and models for increasingly complex mixtures. Analytic equations of state based on the Statistical Associating Fluid Theory (SAFT) allow to model complex systems. The term "complex" here refers to strongly asymmetric mixtures, such as in polymer-solvent mixtures, or to fluids with anisotropic interactions, such as associating and polar substances, or to structurally anisotropic fluids like liquid crystalline materials. Also mixtures with charged species require models with a sufficiently detailed molecular picture. J. Gross also showed that SAFT models can serve in various new application domains. As examples, they are used for integrated solvent and process design, or when applied with the classical density functional theory, these models provide a powerful framework to predict interfacial properties. SAFT-models are also used to speed up molecular simulations, for example by estimating vapour-liquid bias potentials in the grand canonical ensemble or they are used to very efficiently adjust force field parameters. A review and discussion of some recent applications, such as the link between SAFT-models and viscosity prediction, was also given.
The first industrial lecture in this topic was given by G. Folas from Statoil who talked about: "Advanced models in industrial praxis: from process design to process optimization". He started by explaining that Statoil, as many other companies, is mostly using commercial simulators for design, optimization and troubleshooting of production facilities. Initial phases of a project may be executed internally, and by the time that the design concept is frozen, the design is further detailed by engineering companies. The general philosophy is to use standard commercial, widely used tools, to the maximum extent. In order to enhance the capabilities of the company and overcome shortcomings of commercial simulators, either CAPE-OPEN packages or stand alone tools (developed either internally or in collaboration with universities) are being utilized for specific applications. His lecture presented examples of the application of advanced models in process design and process optimization. A range of applications such as the confidence in the calculation of water content of natural gas on the design of processing facilities, evaluation of corrosion risk in production pipelines, gas hydrate inhibition, evaluation of freezing risk in low temperature gas processes and the use of models such as CPA and SAFT were discussed. The next lecture addressed by M. Heilig from BASF dealt with "industrial applications of molecular thermodynamics with focus on screening and extrapolation". His lecture explained that physical property data and thermodynamics are the fundamental basis for process development and chemical engineering applications. The starting point of all design is and will remain in foreseeable future experimental data. A comprehensive physical property database is therefore essential. These data are used as basis for fitting equations which allow a quick access to process relevant data and also to process ideas and concepts. Two aspects were highlighted: thermodynamic screening applications and extrapolation of physical property data. Thermodynamic screening methods can be applied for the selection of process and functional solvents. Converting the process idea or the functional application to physical property criteria, the available data from databases have to be supplemented by predictive modelling for a multitude of components. It was shown that, in addition to group contribution methods, the quantum chemical continuum solvation model Cosmo-RS could be very useful. Extrapolation means the prediction of data in wide ranges of temperature, pressure and concentration as well as the prediction of higher systems, based on pure component and binary data. M. Heilig showed that CPA and PCSAFT-type equations of state enabled a physically more sound extrapolation. However a complex pure component parameterization is required. Molecular simulation, based on intermolecular potentials obtained from fewer thermodynamic data, requires time and effort and is so far only considered for important systems where no other solution exist. Besides a nearly complete set of physical property data molecular simulation enables access to structural information of fluids and fluid mixtures.

Molecular Tools for Data Management
The third plenary lecture, the title of which was: "NIST ThermoData Engine: Increasing Value, Preventing "Pollution", Broadering Scope, and Providing Communications for Thermodynamic Property Information" was given by M. Frenkel from NIST who stated that the NIST ThermoData Engine (TDE) represented the first full-scale software implementation of the dynamic data evaluation concept for thermophysical and thermochemical property data. He explained that this concept requires the development of large electronic databases capable of storing essentially all relevant experimental data known to date with detailed descriptions of relevant metadata and uncertainties. The combination of these electronic databases with artificial intelligence (expert-system) software, designed to automatically generate recommended data based on available experimental and predicted data, leads to the ability to produce critically evaluated data dynamically or 'to order'. The scope of the TDE includes pure compounds, binary mixtures, ternary mixtures and chemical reactions. TDE is a critical component of the Global Information System in Science and Engineering (GISSE) in application to the field of thermodynamics (ThermoGlobe). Role and feasibility of use of a variety of molecular tools (such as chemical system identifiers InChI) and property prediction methods (group contributions, QSPR with SVM technology, Monte-Carlo and ab initio) in critical data evaluation, as well as in global data validation process, involving major journals in the field and supported by TDE, were discussed. Various options for bundling TDE with engineering software applications, including chemical process design engines, were illustrated. In his conclusion, M. Frenkel explained that TDE technology was incorporated into online software to aid the process of experimental planning for property measurements. The software is to be used by experimentalists through open domain free Web access worldwide.
The next lecture was focused on: "Are there ways to improve the accuracy of predictive methods in the field of thermodynamic properties?". It was given both by R.J. Meier and G. Krooshof, both from DSM who started by explaining that thermodynamic data are key in the understanding and design of chemical processes and that there are various ways to obtain such data. They stated that next to the experimental determination, computational methods are extremely valuable in the design of new chemical routes involving new chemical species, and sometimes indispensable tools in obtaining, e.g., heats of formation and Gibbs free energies but also many other quantities. It is meanwhile becoming recognised that the accuracy of the predictive tools is often not sufficient. R.J. Meier noticed for instance that boiling points can be very much off. Chemical accuracy for (free) energies required is about 1-4 kJ/mol. Regarding heats of formation and Gibbs free energies the major toolboxes to obtain such quantities by computation are quantum mechanical methods and group contribution methods. The group contribution methods are common in industrial context. On the other hand, quantum chemical calculations can essentially treat any species, whether it can be broken down into useful groups or not. However, although a lot of progress was made in quantum calculations over the last decade, for the majority of chemical species we are still quite a bit away from what is often referred to as chemical accuracy, i.e. 1 kcal/mol. The last industrial invited lecture was given by M. Kleiber from ThyssenKrupp Uhde GmbH. It was devoted to the "engineering point of view on thermophysical properties". Indeed, in an industrial company, many people are concerned with physical properties. They usually have different backgrounds which make it difficult for them to even communicate with each other. There is a large gap between the specialist for physical properties, who is more or less focused on getting the correct values for a given problem on demand, and the layout engineer, who hardly knows the meaning of the particular properties but has to ensure that every piece of information is transferred correctly. Physical property specialists must not restrict themselves to the determination of physical properties but should get involved in the processes as much as possible, as they might be the only people which are able to detect and solve the various problems encountered by the engineers who are less familiar with the thermodynamic issues. Focusing on a typical engineering company like Uhde, the presentation referred to some of the everyday problems of a process simulation engineer, e.g. the problem with derived properties like liquid state isobaric heat capacity, the model change in a process simulation or the behavior of substances at extreme conditions. It was referred what the preferences are when a model has to be chosen and what the demands are when a new model is established in a company. It was pointed out what the real gaps in the physical property description are. The roles of estimations and the accuracy requirements in process simulation were discussed. It is stressed that the application of estimation methods is not a matter of accuracy but of responsibility. Finally, a perspective was given regarding the role of molecular thermodynamics in an engineering company in the future. According to M. Kleiber, molecular thermodynamics can contribute to the development of new correlation approaches but will never be incorporated in process simulators. The application of molecular thermodynamics will remain restricted to specialists.

Round Table discussion
The round table discussion was the occasion to bring together professionals from various origins. It was moderated by P. Ricoux (TOTAL) and M. Brehelin (Rhodia). The panel was otherwise composed of: -two industrial 'users': O. Koch who represented the engineering company Linde Engineering and P. Pullumbi who represented Air Liquide; -two software vendors: P. Ungerer from Materials Design that provides as well as develops molecular simulation tools and services, and J.-C. Mani from Process Systems Enterprise (PSE), who is at the cross-road between academia and process engineering industry; -and finally two university representatives: P. Sautet from ENS Lyon, who is a specialist in computational chemistry for catalytic applications and A. Padua from Clermont-Ferrand University, who heads a laboratory that has both experimental and simulation activities.
The discussion was constructed around four main questions.
1. Your experience: How do you consider the need and/or the use of new molecular thermodynamics tools (molecular simulations and/or equations of state) in your environment? Are they used in a satisfactory manner? Do you believe a more intensive use would be beneficial? How do you think the working party on Thermodynamic and Transport properties could help?
Obviously, the return of experience depends very much on the type of activities concerned. The most obvious distinction concerns the different visions regarding time and money: for industry, time is a strong limitation, not money; the opposite is true from the academic point of view. Industry aims at solving problems as quickly and efficiently as possible. Various tools can be used to that end (from equations available in commercial process simulators through specific software, including molecular simulation to laboratory data acquisition). They should not be opposed to one another but rather be considered as complementary. A continuing research is needed in order to improve their integration.

Correct Tools for Correct Use
All agree that the use of molecular thermodynamics tools requires some kind of specific expertise. While some companies have hired a specialist whose main responsibility is to develop in-house applications, others consider that the expertise threshold is too large in view of the possible benefits, at least at the present time. Different objectives can be envisaged: -the development of predictive equations of state may take advantage of the possibility offered by molecular simulation that is to isolate specific phenomena, thus fine-tuning the equation to a specific need. This approach allows systematically and rationally improving the representation of the system and transferring molecular-based features to a more coarse-grained model like the equation of state; -the use of equations of state often comes with a specific difficulty which is the adequate parameterization. Here more fundamental models may be useful either to generate pseudo-experimental data in order to complement the databases, or in order to help determining the parameters from their atomistic significance (use of ab initio approaches); -the computational based approaches like COSMO-RS are mainly used for screening purposes. As these methods are based on ab-inito calculations and do not require empirical parameters their use allows to rapidly explore many situations for which few or no data exist; -meso-scale methods which are coarse-grained molecular simulations approaches are another group of methods that are becoming very useful due to the possibility to explore time and length-scales that are of technological reach. These scales cannot yet be explored with atomistic simulation methods.

Process Life-Cycle
Large companies who develop new processes insist on the different approaches depending on the stage in the life-cycle of the process development: -in the early stages, a screening tool is important. Here, accuracy is not that important, and a stand-alone software can be used. Hence, there is a clear position for use of molecular simulation or similar software; -in a second stage, when process-design is involved, the generation of data is important. In this case, the molecular simulation is used in a "production" mode to complement experiments so as to reduce the time needed for generating data. A package that can be interfaced with a process simulator is also essential. High precision is not so much stressed but rather good trends. Group contribution models can very well be used. CAPE-OPEN was not mentioned in the discussion but could be of great help here; -in a final stage, accuracy is essential and a specific model can be used. Most often, because of long term habits and confidence into the ancient correlative models, it will be very difficult to convince process engineers to use new models: they should therefore be confident in the results. The criteria for this very much depends on the type of industry considered (petrochemical, chemical, pharmaceutical, etc.).

Virtuous Triangle
In order to increase the use of molecular simulation in the industry, there is a need of easy to use tools. Software vendors have to be the link between industrial and academic partners. There should be a virtuous triangle between industrials end-users, academic researchers and software vendors. It will be an iterative process to obtain tools matching the needs of industrial users, the vendors being responsible for the software quality, while the academic partners guarantee the theoretical foundations. These privileged relationships should not hide the need of direct contact between industry and academic partners, for the benefit of all. Having a direct vision of future needs, the industrial contribution will help identify research paths, while taking advantage of the knowledge of experts as well as tailor-made software packages. The research groups will thus get funding and sense of purpose for their developments.

Data Gathering Versus Understanding Phenomena
The academic participants stress their effort that is more related to the understanding of the underlying phenomena rather than the production of data. Data gathering is recognized as an important work but difficult to publish as such. It is remarkable that today, data are sometimes shown without referring to the original authors but rather (at best!) to the database they were found in. This disregarding of scientific conduct may not be tolerated by the scientific community; it is a very unfortunate development for experimental groups. True experimental data remain essential for tuning the model parameters. This is why healthy complementarities must be developed between the various fields.
2. Models: What bottlenecks do you see regarding the use of the models (new systems to be investigated, new properties to look at, etc.?). How do you see these tools can be useful for scale-up issues?
How to Choose a Model?
Although some industrial users dream of a single model that works as a black box, most specialists consider this to be unrealistic: it is always possible to hide a complex code behind a userfriendly interface but this will not change the intrinsic limitations. The need for new, improved models is clear due to the increasing complexity of the problems to be solved. The history of molecular-based tool developments during the last two decades indicates that this field is very active and its combination with the extraordinary improvement of the hardware speed and memory will surely be able to treat problems in the future that are not tractable at present.
As a consequence, it is important to remain knowledgeable so that the most pertinent tool can be recommended in view of the need. This is an essential role of the software vendors. It is often not easy to answer as it requires a unified vision, going from the fundamentals to the applications. This issue refers back to the discussion presented earlier regarding the process life-cycle: different approaches must be recommended depending on the development stage of the process.

Parameterization
The discussion regarding models can also be envisioned from the point of view of parameterization: should one develop parameters for specific applications (more accurate) or rather for large applicability? The answer is not universal but leads to the reflection concerning the acceptable accuracy, which can be very variable. This implies: -that it is possible to evaluate the model accuracy, referring back to the availability of data and their own accuracy (which is rarely available!); -that the effect of the thermodynamic model uncertainties on the final process simulation is known; this is also rarely done.
There are many industrial examples where the results of process simulations are sensitive to the specific parameterization used in the models. The choice of the data used for parameterization leads to following questions: which are the good data, which are less reliable data? How shall the parameter fitting be performed? Therefore, the evaluation of thermophysical property data should play an increasingly important role in industrial thermodynamics.

New Models for New Properties
It is also stated that more efforts should be put on the accuracy of calculations for derivative properties (heats of mixing, heat capacities, speed of sound, as well as on models for transport properties, particularly for mixtures, e.g. the viscosity of a mixture of two or more compounds).

Upscaling
The conclusion of this discussion clearly states that the different approaches (equation of state; molecular simulation or even experimental data-gathering) should not be opposed. The tools are fully complementary. Molecular simulation has two interests: help understanding phenomena and generate pseudo-experimental data. Regarding the first point, qualitative results are expected. Simulation is used as an exploratory model making it possible to identify trends and to increase their understanding for new compounds or new areas (biomass, oxygenated compounds, fluorine, etc.). It is then an ideal tool for scale-up approaches: it is essential to investigate new ways to bridge the different time and length-scales. A practical way is to use ab initio tools for parameterization of larger scale models; in the opposite direction, it might be possible to initialize a small-grain model with a larger grain approach.
Concerning the generation of pseudo-experimental data, experiments have to be used to validate results of molecular simulations. Once it is validated, molecular simulation can complement lab experiments when experiments are not easy to obtain (specific pressure and temperature conditions, reactive systems, safety conditions, etc.) or to accelerate the production of data.
3. Algorithms: Considering the increasing complexity of these tools, how do you consider they should be made available? Comment regarding run-time or complex phase equilibrium issues through process simulators or other interfaces. Is there an issue in further investigating parallelization or numeric optimization? How to use best intensive calculation tools for these methods and applications?

Improved Use of Computer Resources
It is important to make an optimal use of computer power (parallelization and numeric optimization) to facilitate dealing with increasingly complex models. This is now also true for the complex equations of state which clearly are more demanding from that respect. Process simulators are not yet ready to investigate multiple core calculations but things will have to change.
This issue is however particularly sensitive for Monte-Carlo or molecular dynamics simulations. In practice, it appears that today's programmers must also become familiar with computer science.
A balance must be found between the optimal use of the new computer architectures, which may require recoding, and the need to capitalize on good-functioning old codes. The answer may be quite different depending on the complexity of the algorithms (solving equations of state requires more complex algorithms than ab initio or molecular simulation calculations).
If recoding is needed, it is important to bring up the question of computing language: most codes available today are still based on Fortran, while new object-oriented languages may be easier to maintain and to interface. The systematic choice of the Windows platform in industry may also be limiting. The final choice will be a compromise between computing time efficiency, development time, maintenance and expected results.
It must however be kept in mind that the most important advances are related to the human factor: i.e. how does the user interpret the results. Common sense may sometimes bring more understanding than many calculations, while paying attention to the false "good intuitions" and the preconceived idea, which can be erroneous because not established at the right scale.
At the end, performances must not mean loss of robustness: industrial users will need "stable applications", which can react intelligently when they are misused.
It was shown that the acceptance of the new developments requires that the industrial users gain confidence in these tools. This can be done through training. Education should be considered at different levels: on the level of the engineering diploma It is essential that the education at this level be broad enough. The basic education in chemistry, physics or chemical engineering is not adequate to provide the majority of the student with a broad overview about the theory, applications, capabilities, potentials and risks of theoretical methods. Even though it might be useful that every student has such a lecture of theoretical physico-chemistry, necessarily limited in scope, most important is that our engineers be aware of the limits of their knowledge and are able to go to a specialist when they need. This is true for theoretical tools, but also for analyzing experimental results: our engineers often lack the basics of metrology.
-on the level of industrial training programs Workshops/training courses make a lot of sense to add specific property know-how ontop of the ChemE education. Regarding molecular simulation, the need from some industrial partners is to have a recent overview about the opportunity and risks of available simulation tools. If it turns out that an expert is needed a team member will be designated (or hired) for this specific mission. Training programs are offered by software vendors, using their proprietary platforms. This is often a very convenient way to find out about the possibilities offered.
-on the level of the education of experts Industry does hire 'fully equipped experts', even though this is on an occasional level. These candidates generally have a PhD in molecular thermodynamics or theoretical chemistry and molecular modelling. Some academic formations exist, such as ATOSIM 2.0 proposed by ENS Lyon, University of Amsterdam and University of Roma. Lifelong training is of course required, through for example conferences, where researchers and industrials meet regularly, or educational workshops and training courses.

SCIENTIFIC CONTRIBUTIONS IN THIS OGST SPECIAL ISSUE
Eight contributions from the conference are presented in this volume. They cover a very wide range of modelling approaches (conventional models, advanced theories, molecular simulation, quantum-chemistry) and applications (oil & gas mixtures, electrolytes, adsorption, etc.). They are discussed below suitably grouped.

Conventional Models
In the terms conventional or standard models are included the classical approaches of applied thermodynamics such as the ones related to cubic equations of state, activity coefficient models and group contribution approaches. These methods are often considered to be mature but are often further developed and/or applied in new ways.
R. Torres, J.-C. de Hemptinne and I. Machin modified the classical method of Grayson-Streed by incorporating an entropic (Flory-Huggins type) term in the regular solution activity coefficient part of the model. Without such a term results for certain size-asymmetric systems are qualitatively wrong. The improved Grayson-Streed approach is shown to perform very well in predicting the hydrogen solubilities in alkanes and heavy oil cuts over a very extended temperature and pressure range.
A. Zaitseva and V. Alopaeus proposed a new approach for improving the estimation accuracy of group contribution methods (for pure compound properties). Their approach, which exploits the chemical similarity of compounds whose properties are estimated, is based on what is called distance weighted optimization, as opposed to the non-weighted optimization often used. The proposed methodology is generic and can be applied to different group contribution methods. The authors have successfully applied their approach to the well-known Joback-Reid and Marrero-Gani methods for the normal boiling point, as well as for several more properties exclusively for the Joback-Reid method. In all cases, the estimations are improved compared to the original methods. The new approach may be even for more useful for group contribution methods employing second and third order groups.
The last contribution in this section is that by H. Dardour, P. Ce´zac, J.M. Reneaume, M. Bourouis and A. Bellagi who investigated an absorption-diffusion refrigeration system using three working fluids, nonane as absorbent, propane as refrigerant and hydrogen as the inert auxiliary gas. The authors performed design simulations of this system using the commercial simulator ASPEN Plus with the Peng-Robinson equation of state as the property prediction model (which is preferred over the Chao-Seader method from the authors). Heat and mass characteristics have been analyzed as well as performances for a wide range of operating conditions and design parameters. While further work may be needed to verify these conclusions, the results point out that the studied nonane/propane/hydrogen system obtains good cooling performances with low generator temperature and it may thus be a suitable alternative to the ammonia/water/hydrogen working fluid which is widely used in commercial cooling units. The authors considered both a version of the model using salt-specific and one with ion-specific parameters, with the former being overall better. The model represents satisfactorily, with few adjustable parameters, vapour pressures, osmotic and mean ionic activity coefficients for 13 1:1 aqueous salt solutions over a temperature range up to 363.15 K. Some good predictions for vapour pressures of two-salt mixtures were also shown.

Molecular Simulation and COSMO-RS
Molecular simulation and computational quantum chemistry approaches have become very popular over the last 20 or so year and for diverse applications.
J. Janecek studied, in a theoretical work, the influence of periodic boundary conditions on the fluid structure and on the thermodynamic properties as computed from molecular simulations. This work is essential in view of investigating the effect that the various molecular parameters may have on different scales. He concludes that when thermodynamic properties are calculated using sufficiently large systems, the effect of local anisotropy, can be neglected.
O. Toure, C.-G. Dussap and A. Lebert compared systematically the predictive capability of COSMO-RS to two other methods (ChemAxon, ACD/Labs) for pKa values of about 50 amino-acids, dipeptides and tripeptides. Such information is useful in food science/industry and the authors find that COSMO-RS, while being somewhat inferior to the other two methods, is overall a powerful truly predictive approach for pKa calculations.

Adsorption and Molecular Simulation
Adsorption phenomena are complex and general reliable predictive methods are not available, especially for multicomponent adsorption. Molecular simulation can provide very useful input and can also contribute to better understanding of the underlying mechanisms. Two articles in this volume deal with molecular simulation applications to adsorption.
X. Rozanska, P. Ungerer, B. Leblanc and M. Yiannourakou combined periodic DFT (Density Functional Theory) with Monte-Carlo methods for the description of adsorption phenomena in complex systems. DFT provides input parameters (electrostatic potentials of adsorption materials) which can be directly used for the molecular simulations of fluid adsorption isotherms, where reliable force field parameters are needed. In this way, complex solids can be studied via simulation without a priori knowledge of the electrostatic parameters. The authors have used their approach to various solids and they have evaluated their affinity for the adsorption and separation of CO 2 /N 2 mixtures in presence of water. Both excess adsorption isotherms and heats of adsorption have been calculated in all cases (single and mixed gas adsorptions).
In the second work, J. Puibasset illustrates how the molecular simulation models used for adsorption in porous materials can be significantly improved by explicitly considering the interdependence between domains. In this way, it is possible to obtain better characterization of porous materials compared to the simpler approaches often used where the porous domains are considered to be independent. J. Puibasset proposes a multiscale approach which allows for molecular fluid/fluid and fluid/substrate interactions as well as for the connectivity between the various domains of the porous material. The usual periodic boundary conditions applied to each domain are replaced by appropriate explicit boundaries. The improvement with the new method is especially pronounced in the high pressure/hysteresis region.

CONCLUSIONS
The workshop has been an excellent occasion for exchanges between experts in molecular thermodynamics from different origins. As an activity sponsored by the EFCE working party on "Thermodynamics and transport properties", it can be considered as a successful step towards creating a community in this field with active participation from industry.
Several points have been brought up that may require future work: -a large number of methods exists, whose pertinence very much depend on their final use.
There is clearly no competition between the various tools but they should rather be seen as complementary. It may be of use to propose a review of the latest developments along with the best ways to employ the complementarities between tools; -very little reflection is put today in the investigation of the complex links that exist between the quality of the experimental data used for parameterization and the uncertainties resulting from the final results. This leads to a certain devaluation of the high quality experimental work that must be recognized for its true value. What about the use of molecular simulation tools for producing pseudo-experimental data, and as a consequence the introduction of these data in the usual commercial databases? -Further work is needed in order to develop methodologies for upscaling the computational approaches: how to navigate among the different scales from quantum to molecular, mesoscopic or bulk thermodynamic models?
Although the number of scientific meetings should remain low, the need for a forum where industrial partners can express their needs, even on a rather basic level, is essential. The working party should consider creating opportunities for such meetings in a regular fashion: they make it possible to bring up questions and needs.