Arxiv:1902.06219V2 [Cond-Mat.Soft] 6 May 2019 Membrane That Allows Only the Solvent to Pass While Retaining So- Lutes

Journal Name

Osmosis, from molecular insights to large-scale applications

a,† a, Sophie Marbach and Lydéric Bocquet ∗

Osmosis is a universal phenomenon occurring in a broad variety of processes and fields. It is the archetype of entropic forces, both trivial in its fundamental expression – the van ’t Hoff perfect gas law – and highly subtle in its physical roots. While osmosis is intimately linked with transport across membranes, it also manifests itself as an interfacial transport phenomenon: the so-called diffusio-osmosis and -phoresis, whose consequences are presently actively explored for example for the manipulation of colloidal suspensions or the development of active colloidal swimmers. Here we give a global and unifying view of the phenomenon of osmosis and its consequences with a multi-disciplinary perspective. Pushing the fundamental understanding of osmosis allows one to propose new perspectives for different fields and we highlight a number of examples along these lines, for example introducing the concepts of osmotic diodes, active separation and far from equilibrium osmosis, raising in turn fundamental questions in the thermodynamics of separation. The applications of osmosis are also obviously considerable and span very diverse fields. Here we discuss a selection of phenomena and applications where osmosis shows great promises: osmotic phenomena in membrane science (with recent developments in separation, desalination, reverse osmosis for water purification thanks in particular to the emergence of new nanomaterials); applications in biology and health (in particular discussing the kidney filtration process); osmosis and energy harvesting (in particular, osmotic power and blue energy as well as capacitive mixing); applications in detergency and cleaning, as well as for oil recovery in porous media.

1 Introduction to counteract the ﬂow: the applied pressure is then equal to the From the etymological point of view, osmosis denotes a “push” osmotic pressure. and indeed osmosis is usually associated with the notion of force and pressure. Osmosis is a very old topic, it was ﬁrst observed centuries ago with reports by Jean-Antoine Nollet in the 18th cen- solute tury. It was rationalized more than one century later by van ’t hydrostatic pressure drop Hoff, who showed that the osmotic pressure took the form of a perfect gas equation of state. In practice, an osmotic pressure relaxation is typically expressed across a semi-permeable membrane, e.g. a

arXiv:1902.06219v2 [cond-mat.soft] 6 May 2019 membrane that allows only the solvent to pass while retaining solutes. If two solutions of a liquid containing different solute concentrations are put into contact through such a semi-permeable membrane, the fluid will undergo a driving force pushing it to- semi- permeable membrane wards the reservoir with the highest solute concentration, see Fig. 1. Reversely, in order to prevent the fluid from passing Fig. 1 : Key manifestation of osmosis. A semi-permeable membrane through the membrane, a pressure has to be applied to the fluid allows transport of water upon a solute concentration difference (in red). The flow of water is directed from the fresh water reservoir to the concentrated reservoir.

a Laboratoire de Physique , Ecole Normale Supérieure, PSL Research University, 24 rue Osmosis is therefore extremely simple in its expression. Yet it Lhomond, 75005, Paris † now at Courant Institute, New York University, New York is one of the most subtle physics phenomenon in its roots – it re- 1,2 ∗ Corresponding author e-mail: [email protected] sulted in many debates over years . Osmosis also implies subtle

Journal Name, [year], [vol.],1–43 | 1 phenomena, in particular as a prototypical illustration for the ex- ney, blue energy harvesting, etc. We conclude with some final, plicit conversion of entropy of mixing into mechanical work. In brief, perspectives. spite of centuries of exploration, osmosis as a field remains very lively, with a number of recent breakthroughs both in its concepts 2 Osmosis : the van ’t Hoff legacy and applications as we shall explore in this review. A simple rea- son for the importance of osmosis is that it is a very powerful 2.1 A quick history of osmosis phenomenon: giving just one illustrative number, it is amazing to We start this review with a short and non-exhaustive journey realize that a concentration difference of 1.2 molar, which cor- through time in order to highlight how a complete understand- ∼ responds roughly to the difference between sea and fresh water ing of osmosis emerged over time. We refer e.g. to Ref. 4 for a (and can be easily achieved in anyone’s kitchen), yields an os- more detailed historical review. The first occurence of the term motic pressure of 30 atmospheres. This is the hydrostatic pres- "osmosis" and clear observation of its effects – beyond the semi- ∼ sure felt under a 300m water column ! Osmosis has potentially nal work of Nollet – is reported at least as early as in the works of a destructive power, in particular in soft tissues and membranes, Henri Dutrochet in the 1820s 5,6. He observed swelling events with possible fatal consequences 3. This explains actually why it or emptying of pockets driven by the presence of various dis- is also an efficient asset for food preservation (such as fish and solved components in water (different sugars in plants, sperm meat curing with dry salt). in slugs...). In reference to the greek term "osmose" (meaning Osmosis is accordingly also a key and universal phenomenon "impulsion" or "push") he introduced the vocabulary "endosmose" occurring in many processes, ranging from biological transport and "exosmose". Interestingly, Dutrochet served as a pioneer in in plants, trees and cells, to water filtration, reverse and forward linking these different topics by claiming that the same physical osmosis, energy harvesting and osmotic power, capacitive mixing, force could be used to describe all these events 5, which is indeed oil recovery, detergency and cleaning, active matter, to quote just a unique and fascinating feature of osmosis. Yet, the mechanisms a few. driving osmotic flow were still unclear, and entangled (or believed The litterature on osmosis and its consequences is accordingly to be entangled) with capillary and electrical effects. In 1854 T. absolutely huge ∗, and it may seem hopeless to cover in a single Graham introduced the word "osmosis" building on the work of review all aspects of the topic with an exhaustive discussion of Dutrochet 7. all possible applications. Also, such a comprehensive list would probably be useless for readers who want to catch up with the top- (a) ics related to osmosis. In writing this review, we thus decided to rather present a tutorial and unified perspective of osmosis, obviously with personal views, avoiding exhaustiveness to highlight a number of significant questions discussed in the recent literature. The review will therefore explore the fundamental foundations of osmosis, emphasizing in particular the – sometimes subtle – mechanical balance at play; then report on more recent concepts (b) and applications related to osmosis which - in our opinion - prove promising for future perspectives. We will accordingly put in context phenomena like diffusio-osmosis and -phoresis, as well as “active” (non-equilibrium) counterparts of osmosis, which were (l) (r) realized lately to play a growing role in numerous applications in filtration and energy harvesting. Fig. 2 : Osmosis versus diffusion. (a) situation where motion of the The review is organized as follows. We start with some basic red solute particles is governed by diffusion alone (b) osmosis situation, reminder of the fundamentals of osmosis in terms of equilibrium where motion of the blue solvent particles is driven by osmosis; the so- and non-equilibrium thermodynamics of the underlying process. lute particles being "repelled" by the membrane, the membrane exerts We further highlight simplistic views clarifying the mechanical as- an effective force on the liquid (solute + solvent) that drives solvent flow pects of osmosis. We then discuss membrane-less osmosis and the towards the highly concentrated reservoir. so-called diffusio-osmotic flows. We then show how such phenomena may be harnessed to go beyond the simple views of van ’t Interestingly, the distinction between osmosis and pure diffu- Hoff. We then explore the transport of particles under solute gra- sion – without a membrane, see Fig. 2 – is not clear from the be- dients, diffusio-phoresis, and discuss how this phenomenon can ginning. The confusion will grow stronger with the work of Adolf be harnessed to manipulate colloidal assemblies. And we finally Fick in 1855 8, where he claims that diffusive motion (Fickian illustrate a number of applications for the introduced concepts, diffusion) is the driver for osmotic flow (the water concentration from desalination, water treatment, the functioning of the kid- imbalance between the two compartments drives the water flow). The question of finding whether osmotic flow is diffusion-driven or not will be an ongoing debate for a century. That diffusion

The word “osmosis” in Web of Science results in tens of thousands of referenced alone cannot account for osmosis is not widely appreciated. In ∗ papers on this topic. 1957, the debate is deﬁnitely closed by an experimental visualiza-

2 |Journal Name, [year], [vol.], 1–43 tion of water flow, using radioactively labeled water molecules 9 all these theories, two of them caught a lot of attention. One of and verified in Ref. 10. The flows measured were significantly them was the proof of van ’t Hoff’s law using the kinetic theory of higher than that expected by pure diffusion. gas to describe the two solutions 15 (which was later improved for multicomponent systems 16). The other one is acknowledged to- (a) (b) 400 day as the common description of osmosis, and makes use of the water with sugar T = 16,1° 300 T = 13,5° concept of chemical potentials first introduced by Josiah Willard Height in the 17 column porous earthenware Gibbs (actually introduced as a physical descriptor required to measures the pot with copper 200 pressure drop ferrocyanide understand osmosis), that we recall in the next section. precipitate 100 Pressure height (ctm.)

water 0 0 2 4 6 Sugar mass fraction (%) 2.2 Thermodynamic equilibrium

Fig. 3 : First measurements of osmotic pressure.(a) Schematic of a We start with the thermodynamic derivation of the osmotic pres- Pfeffer cell, the device used by W. Pfeffer to perform the first measure- sure as proposed by Gibbs. We follow here the clear-cut presenta- ments of osmotic pressure; (b) Osmotic pressure as a function of solute 18 concentration at two temperatures, with experimental data points from W. tion proposed in the textbook by Callen , which we recall here Pfeffer 11 verifying a linear relation (the lines are a guide for the eye). for the purpose of settling properly the foundations. In addition to Callen it is also worth reading the rigorous thermodynamic treat- 19 In 1877, Wilhelm Pfeffer made the first measurements of os- ment by Guggenheim . We consider a composite system made motic pressure 11, see Fig. 3. At equilibrium, he measured a rise in of two simple reservoirs (left and right) separated by a rigid wall the concentrated solution, corresponding to a hydraulic pressure permeable to component w (usually the solvent) and totally im- drop that is equal to the osmotic pressure. He measured a linear permeable to all other components (labelled s1, s2 and so on). relation between the osmotic pressure and the concentration dif- The whole system is in contact with a thermal bath at tempera- ference. But also, Pfeffer measured that for each degree rise in ture T. The solvent is in equilibrium over the whole system, i.e. temperature, the pressure would go up by 1/270 12. This fact was over the two reservoirs, while solutes cannot equilibrate between reported to Jacobus Henricus van ’t Hoff by the botanist Hugo de the reservoirs. Due to the imbalance of solute fraction between Vries and van ’t Hoff immediately recognized that 270 was an ap- both reservoirs, the solvent cannot keep a homogeneous pressure proximation of 273 K. Intrigued by this result, he attempted in across the two reservoirs while ensuring the equality of chemical 1887 to rationalize this linear dependence 13 and suggested to in- potential at equilibrium. An (osmotic) pressure drop builds up, terpret that the osmotic pressure ∆Π was exerted by the solute which the membrane withstands. particles and equal to the partial pressure that they would have Assuming that the solute concentration is dilute, the Gibbs free in gas phase (therefore the term "osmotic pressure"): energy of a binary system of solvent and dilute solute can be writ- [...] it occurred to me that with the semipermeable bar- ten as rier all the reversible transformations that so materially G(T, p,N ,N ) =N µ0 (p,T) + N µ0(p,T) ease the application of thermodynamics to gases, become w s1 w w s s (2) equally available for solutions... That was a ray of light; Nw Ns + NwkBT ln + NskBT ln and led at once to the inescapable conclusion that the Nw + Ns Nw + Ns osmotic pressure of dilue solutions must vary with tem- 0 0 perature entirely as does gas pressure [...]. 12 where µw(p,T) and µs (p,T) are the chemical potentials of the pure solvent and solute and the last two terms correspond to the then writing entropy of mixing terms. In the dilute regime where Ns Nw the ∆Π = kBT∆cs (1) Gibbs free energy simplifies to with k the Boltzmann constant, T temperature and ∆c the solute B s 0 0 Ns G(T, p,Nw,Ns) = Nwµw(p,T) + Nsµs (p,T) kBTNs + NskBT ln imbalance between reservoirs. − Nw Eq. (1) is today referred to as the van ’t Hoff law, and gives in (3) practice good agreement for the osmotic pressure measured be- and the chemical potential of the solvent may be obtained as µw = tween two solutions separated by a membrane permeable only ∂Nw G 0 to the solvent. For a solute imbalance of ∆C 1 2 mol.L 1 µw(T, p,X) µw(p,T) kBTX. (4) s = . − ' − (corresponding to the ionic strength difference between fresh with X Ns/Nw the solute molar fraction. The chemical potential and sea water, which is twice – two ions for salt – the typi- ' (l) (r) balance, µ = µ , thus writes 1 w w cal concentration 0.6 mol.L− ), we find an osmotic pressure of (l) (r) ∆Π = kBT∆CsNA 30 bar. Ns Ns ' µ (T, p(l),0) k T = µ (T, p(r),0) k T . (5) At the time, the interpretation of van ’t Hoff gave rise to a num- w B (l) w B (r) − Nw − Nw ber of debates 1,2. In the following decades a great number of the- (r) ories were invented to describe the osmotic phenomenon and a Noting then that for small pressure drops, µw(T, p ,0) (l) (r) (l) ' detailed review of these theories can be found in Ref. 14. Among µw(T, p ,0) + (p p )vw, with vw = ∂pµw(T, p,0) the molec- −

Journal Name, [year], [vol.],1–43 | 3 ular volume, one deduces finally with a thermal bath at temperature T. The entropy production (per unit membrane area ) is accord- " (r) (l) # A (r) (l) Ns Ns ingly written as: ∆Π = p p = kBT (6) − V (r) − V (l) (r) (r) T dS (r) (l) dNw (r) (l) dNs Φ = = µw µw µs µs (10) Introducing the concentration as cs = Ns/V, one thus recovers the A dt − − dt − − dt result of van ’t Hoff dN(r) k T c i ∆Π = B ∆ s. (7) with dt the flux of molecules of component i per unit area. The (r) dNi In the case of several dilute solutes, this generalizes simply to dissipation function of Eq. (10) is a product of fluxes dt and the corresponding thermodynamic forces, here the differences in " (r) (r) (l) (l) # Ns1 + Ns2 + ... Ns1 + Ns2 + ... chemical potentials. ∆Π = kBT . (8) V (r) − V (l) Now, restricting ourselves to ideal solutions for simplicity, one (r) (l) may write the chemical potential difference as µ µ = vi∆p+ The derivation above is limited to dilute solutes. For arbitrary i − i kBT∆lnXi where Xi is the molar fraction of component i and molar fractions X of solute/solvent mixtures, the osmotic pres- (r) (l) vi = (∂ µi/∂ p) the molar volume of i. Accordingly, µs µs = − sure is given in terms of the general expression for the pressure, ∆cs (r) (l) ∆cs vs∆p + kBT for the solute and µw µw = vw∆p kBT for namely 20 cs − − cw ∂ f the solvent (where we used cs cw). Eq. (10) then rewrites: Π(X) = X f [X] + f [X = 0], (9) ∂X − (r) (r) ! (r) (r) ! dNw dNs 1 dNs 1 dNw with f (X) = F/V the Helmoltz free energy density calculated for Φ = vw + vs ∆p kBT∆cs − dt dt − cs dt − cw dt a solute molar fraction X. Deviations from ideality are for exam- (11) ple measured for polymers, where the range of validity of the van From the dissipation function in Eq. (11), we may thus identify a ’t Hoff law decreases with increasing molecular weight 20. Devia- new set of forces and fluxes: new forces are ∆p and kBT∆cs, tions are also expected for highly concentrated brines or solvent − − respectively the hydrostatic pressure and solute concentration im- mixtures, e.g. in the context of solvophoresis, see below 21. balance; new flows are (a) the total volume flow through the An interesting, and quite counter-intuitive remark is that – pro- membrane (sum of all flows): vided it is semi-permeable – the membrane characteristics do not (r) (r) appear in this thermodynamic expression for the osmotic pres- dNw dNs Q = v + v (12) sure. Another puzzling remark is that the osmotic pressure is a w dt s dt colligative property, i.e. it does not depend on the nature of the so- and (b) the excess solute flow (as compared to the solute flow lute (nor that of the membrane), but only on the concentration of carried by the solvent) or the exchange flow: the solute. This is relevant when the membrane is completely im- (r) (r) permeable to the solute, but when the membrane is only partially 1 dNs 1 dNw Je = (13) impermeable, or when there are different solutes with different cs dt − cw dt permeation properties, there may be both a solvent and a solute flux driven by the solute concentration imbalance (in opposite Under the assumption that the concentration of solute is small 22–26 cs cw, one may thus rewrite Je Js/cs Q where Js is the solute directions) . The osmotic pressure is then usually assumed ' − to be reduced by a so-called (dimensionless) reflection factor, σ, flow. which depends on the specific properties of solvent-membrane in- The framework of irreversible processes assumes a linear rela- 30 teractions and transport. This requires to go beyond the thermo- tion between fluxes and forces , hereby taking the form dynamic equilibrium and consider the detailed mass and solute ! ! Q ∆p transport across the membrane, as we now explore. = L − . (14) Js csQ × kBT∆logcs − − 2.3 Osmotic fluxes and thermodynamic forces where L is the transport matrix. Importantly, as we discuss below Following the work of Staverman 27, Kedem and Katchalsky and in Sec. 3.2.2, this matrix is symmetric according to Onsager’s derived a relation between solute and solvent flows through principle – due to microscopic time reversibility – and definite a porous membrane and the corresponding thermodynamic positive – due to the second principle of thermodynamics –. forces 28, based on Onsager’s framework of irreversible pro- The question then amounts to characterizing the transport co- cesses 29,30. efficients of this matrix. By identifying limiting regimes, Kedem As in the previous section, we consider a composite system and Kachalsky rewrote these transport equations in a more ex- 28,31–34 made of two simple reservoirs (left and right), containing a sol- plicit form as vent w and a solute s. The reservoirs are separated by a rigid wall, Q = L (∆p σkBT∆cs), (15) which is now permeable to all components, but with a differential − hyd − permeability between the solute and the solvent. Obviously, the Js = LDωs∆cs + cs(1 σ)Q, (16) objective of the membrane is somehow to reject the solute but the − − rejection is incomplete here. The whole system is put in contact where Lhyd = κhydA /(ηL) is the solvent permeance through the

4 |Journal Name, [year], [vol.], 1–43 membrane with κhyd the permeability (with units of a length The specificity of the membrane and its interaction with the squared), A the membrane area, η the fluid viscosity, and L solute molecules actually come into play into this reflection co- the membrane thickness; LD = A Ds/L is the solute permeabil- efficient σ. A number of models have tried to rationalize the ity with Ds the diffusion coefficient of the solute. Eq. (15) is often dependence of σ on the chemical and physical properties of the referred to as the Starling equation in the physiology literature 35, components. The first models took into account steric effects (sim- see e.g. Refs. 36,37. The osmotic pressure generated by the large ilar to Fig. 4), where in fact the volume accessible to the solute scale molecules involved in the body (complex proteins such as inside the pore would differ (because of its typically larger size) albumin and more) is referred to as the oncotic pressure. These than that accessible to the solvent 42–44. The next generation of equations introduce two dimensionless (numerical) factors: σ is models sought to include as well hydrodynamic interactions, in- the so-called reflection or selectivity coefficient and ωs is a solute vestigating how friction induced by the proximity of the solute “mobility” across the membrane – both of which we discuss in to the pore walls would reduce permeability 45,46. Anderson also details below. studied interactions with the pore walls and adsorption of the so- 38 The Onsager symmetry relations for Eq. (14) can be veri- lute in the pore . Similarly, the “mobility” coefficient ωs entering fied by exploring two limiting cases: (1) The situation where the transport equations will depend both on the solute-membrane ∆p = 0 yields osmotic flow only as Q = σLhydcs∆µ (using ∆µ = interactions and transport parameters. A first, naive, estimate is kBT∆cs/cs in the dilute case); (2) and the situation where ∆µ = 0 to identify this coefficient with the partition coefficient of the so- yields Js csQ = σLhydcs∆p. One obtains therefore [Q/∆µ]∆p=0 = lute between the membrane and the reservoirs at equilibrium, − m bulk 31 [(Js csQ)/∆p] and the symmetry of the transport matrix is Ks = c /c , so that ωs = Ks . But this estimate does not ac- − ∆µ=0 s s indeed verified. count for the complex transport processes occurring within the The reflection coefficient and the solute mobility – The Kedem– membrane. Interestingly, the non-dimensional coefficients ωs and 31 Katchalsky equations introduce the reflection coefficient σ men- σ are expected to be linearly related, as 1 σ ∝ ωs, a result that − tioned previously and first described by Staverman 27. This coef- we will recover below in a specific case. ficient a priori depends on the relative interactions of the mem- Altogether, a complete determination of the reflection and mo- 22,23,38 brane with the solute and solvent . The Kedem-Katchalsky bility coefficients requires to implement a microscopic description framework also introduces the permeability of the solute through of the membrane-fluid interactions. We will explore below and in the membrane via the combination LDωs. A fully semi-permeable Sec. 3 various situations highlighting how playing with interac- membrane corresponds to the case where σ = 1 and ωs = 0: the tions may lead to advanced osmotic transport behavior. solute flux vanishes Js = 0 and the pressure driving the fluid identifies with the van ’t Hoff result ∆Π = kBT∆cs. Reversely, a “trans- 2.4 Mechanical views of osmosis: a tutorial perspective parent” membrane which is fully permeable to both solute and solvent correspond to σ = 0 and ωs = 1: no osmotic pressure is Beyond the general formalism introduced above, it is interesting expressed and the solute flux reduces to Fick’s law. to get further fundamental insights into the microscopic mecha- In the intermediate case, the membrane is partially permeable nisms which underlie osmosis. In particular it is of interest to get to the solute and we expect 0 < σ < 1, see Fig. 4. As an example, some intuition on the mechanical force balance associated with in a pure nafion membrane about 18µm thick, the reflection coef- the osmotic pressure. To do so, we will reduce the microscopic ficient between water and KCl salt was measured as σ = 0.82 (at ingredients of osmosis to their minimal function and this descrip- concentration 0.25 mol/L) 39. tion has merely a tutorial purpose. Still it is very enlightening in order to understand how the connection between “microscopic” (a)σ = 1 (b) 0 < σ < 1 (c) σ = 0 parameters and thermodynamic forces builds up. Such mechanistic views of osmosis also allow to envision advanced osmotic phenomena, beyond the van ’t Hoff perspective. Alternative approaches with similar illustrative objectives were proposed for one-dimensional single file channels, see Refs. 47,48. We pointed out above that the van ’t Hoff law for the osmotic Fig. 4 : Examples of reflection coefficient based on steric exclusion. pressure does not involve the membrane properties per se, pro- In (a) the blue solvent only may traverse the pores; while in (b) the red vided that it is semi-permeable. So it is tempting to replace the solute particles may also traverse; their permeability through the pores is membrane by a crude equivalent, namely an energy barrier act- however small since the accessible volume in the pore for the red solute particles is smaller than that for the blue solvent. (c) The membrane is ing on the solute only, say U (x) (assuming for simplicity a uni- now fully permeable to all species, and therefore diffusion dominates and dimensional geometry) – see Fig. 5. This approach, which cap- solute particles move towards the low concentration side. tures the minimal ingredients at play in osmosis, was first introduced by Manning 49 in the low concentration regime, and gen- Interestingly, cases with negative reflection coefficient, σ < 0, eralized more recently to explore the osmotic transport across were reported. This situation is often termed anomalous osmo- perm-selective charged nanochannels 50 or in non-linear regimes sis 40,41 and it corresponds to situations where the solute is more at high solute concentrations 51. One may note that such a poten- permeable than the solvent. We will discuss in Sec. 4 various tial barrier can also be physically achieved; for example, it may be examples where such a situation with reversed osmosis occurs. generated from a nonuniform electric field acting on a polar so-

Journal Name, [year], [vol.],1–43 | 5 lute in a nonpolar solvent 52, or it can represent the nonequivalent (a) interactions of solute and of solvent particles with a permeable membrane, e.g., charge interactions 50,53. Let us ﬁrst consider the ideal case where the barrier’s maximum is high, i.e. Umax kBT, so that the solute cannot cross the barrier: this is the perfectly semi-permeable case. In both reservoirs the solute is at equilibrium and the solute proﬁle follows accordingly the Boltzmann relation

(r)/(l) (r)/(l) U (x) cs (x) = cs e− kBT (17) (b) × Now, a key remark is that the force on a ﬂuid element of volume dτ (consisting of the solvent and solute mixture) will write

(r)/(l) d f (x) = cs (x) ( ∂xU (x))dτ. (18) (x) × − U with dτ = A dx; A is the membrane area. The total force per unit area acting on the ﬂuid is accordingly integrated over x

Z ∞ U (x) Z 0 U (x) FT (r) (l) x = dxcs e− kBT ( ∂xU (x))+ dxcs e− kBT ( ∂xU (x)) A 0 × − ∞ × − − (19) L (where we arbitrarily put x = 0 at the position of the maximum of the energy barrier), leading immediately to Fig. 5 : The physical force driving osmosis is interaction of the solute with the membrane. (a) The membrane exerts a repulsive force FT (r) (l) (red arrows) on the solute particles (red) that creates a pressure gradient = kBT [cs cs ] ∆Π (20) A × − ≡ (or a void on the immediate left hand side of the membrane) that drives the flow (blue arrows of blue solvent particles). (b) Mechanical view of where we neglected terms behaving as exp[ Umax/kBT]. Alto- − osmosis: the partially permeable membrane may be viewed as an energy gether this simple approach allows one to retrieve the van ’t Hoff barrier for the solute molecules (in red) that they have to overcome in law. It highlights the mechanical origin of osmosis: as is trans- order to traverse the membrane. parent from the previous derivation, the osmotic pressure results from the fact that the reservoir containing more solute particle Pe = vxL/Ds 1, such that the convective term of Eq. (21) is will generate a higher repelling force on the fluid than from the other reservoir: accordingly a fluid flow will be generated from negligible. This is valid for low permeability (nanoporous) mem- the low to the high concentrations, hence diluting the more con- branes. The full derivation including the convective term was centrated reservoir. considered in Ref. 49. Since the solute flux across the membrane While the above approach is instrinsically at equilibrium, it can Js is constant in time and spatially uniform, Eq. (21) is explicitly be easily generalized to a non-equilibrium situation by releasing solved with respect to the concentration as: the assumption of an (infinitely) high energy barrier: in this case R L/2 r dx exp[+βU (x )] c x c( ) ∆c e βU (x) x 0 0 (22) the solute can cross the “membrane” between the two reservoirs s( ) = s s − L 2 , − R / dx exp[+β (x )] at a finite rate, see Fig. 5, generating a solute flux. We further L/2 0 U 0 − assume that the membrane is fully permeable to the solvent (no where β = 1/kBT. The solute concentration difference between energy barrier acting on it), with a permeance Lhyd relating the (r) (l) the two volumes is ∆cs = cs cs . For simplicity we assumed fluid flux Q to the pressure drop ∆p in the absence of a concen- − that the barrier has an extension L. tration difference: Q = L ( ∆p). hyd − Now turning to the momentum conservation equation for the The stationary dynamics of the system is described by the cou- fluid (solvent + solute), the flow field v of the fluid obeys a Stokes pled set of equations for the solute diffusive dynamics – Smolu- equation (neglecting inertial terms) chowski equation – and fluid transport – Navier-Stokes equation. In the 1D geometry described above, the stationary solute con- 0 = ∇p + η∇2v + f , (23) − ext centration cs(x) obeys a Smoluchowski equation: where p is the fluid pressure and f ext represents the total volume 0 = ∂t cs = ∂x js forces acting on the system, e.g. the forces acting on the solvent − and on the solute, here = ∂x ( Ds∂xcs + λscs ( ∂xU ) + vxcs), (21) − − − f ext = cs( ∇U ). (24) where js = Js/A is the solute flux per unit surface, Ds is the so- − lute diffusion coefficient, λs = Ds/kBT the mobility and vx the lo- The driving force inducing the solvent flow along the x axis cal fluid velocity. We will further assume a low Péclet number, is accordingly written in terms of an apparent pressure drop,

6 |Journal Name, [year], [vol.], 1–43 ∂xP = ∂x p + cs(x)( ∂xU ). The membrane, via its potential related here to the reflection coefficient as ωs = 1 σ. − − − − U , will therefore create an average force on the fluid, which writes per unit surface In this first part we have reviewed the basic understanding of Z L/2 osmosis, from the historical discovery of the phenomenon to the ∆P = ∆p + dxcs ( ∂xU ) ∆p + σ∆Π, (25) precise understanding of the effect in terms of thermodynamic − − L/2 − ≡ − − forces. Although simplistic, the previous mechanical/kinetic ap- where ∆ means the difference of a quantity between the two sides. proach provides a fruitful and complementary perspective on os- The second term of Eq. (25) can be interpreted as the osmotic motic transport, which suggests a number of generalizations – contribution. Using the expression for the concentration profile that we will discuss below. It also reveals that the key aspect of given in Eq. (22), one recovers the classical van ’t Hoff law of osmosis is not really the membrane itself, but the existence of dif- the osmotic pressure, ∆Π = kBT∆cs, and furthermore obtains an ferential forces acting separately on the solvent and the solute. expression for the reflection coefficient σ as This is crucial to understand a number of phenomena related to L osmosis that we discuss below. σ = 1 . (26) − R L/2 dx x L/2 0 exp[+βU ( 0)] 3 Osmosis without a membrane − Situations where differential forces act on the solvent and the The above result correctly recovers the case of a completely solute occur naturally, especially at interfaces: for example a semi-permeable membrane (no solute flux across the membrane), charged surface does act specifically on dissolved ions, repelling i.e., βU 1 and σ 1, yielding ∆P = ∆[p Π]. In the inter- → − − − co-ions and attracting counter-ions; or a neutral hard wall will mediate cases, although the membrane is permeable, a flow arises repel polymers via excluded volume. As we now discuss in the due to the solute concentration gradient even in the absence of an following sections, these specific forces may be harnessed to in- imposed pressure gradient. When the potential is repulsive and duce interfacially-driven osmotic flows. small U kBT, then 0 < σ < 1; the flow is in the direction of ∼ increasing concentration. The geometry we will consider here involves a solid surface along which a solute gradient, or more generally a thermody- Integrating Eq. (23) over the membrane area (A ) and thick- namic force – an electric field, a temperature gradient... – is es- ness (L) allows the total flux Q to be expressed as tablished, as sketched in Figs. 6-7. Under an electric field, the net electric forces occurring within the diffuse interface close to the Q = Lhyd (∆p σkBT∆cs). (27) − − solid will push the fluid and generate a so-called electro-osmosis Here the permeance Lhyd can be expressed in terms of the per- flow for the solvent. But as we will show below, a solute gradient A κhyd meability, κhyd, as Lhyd = L η . The permeability κhyd is de- ∇c∞ parallel to the surface can also generate fluid motion whose 1 2 amplitude is proportional to ∇c : fined formally in terms of the flow as κhyd− = ∇ v / v , where ∞ 1 RR h− i h i = V − dxdA ( ) denotes an average over the pore volume, h·i · vDO ∝ ∇c∞. (29) here V = A L. These parameters, κhyd and Lhyd, take into account the detailed geometry of the pores in the membrane (pore This latter phenomenon is usually coined as diffusio-osmosis. The cross section, length, etc.). Overall Eq. (27) agrees with the phenomenon bears some fundamental analogy with Marangoni Kedem–Kachalsky result in Eq. (15). While this approach is de- effects where a gradient of surface tension at an interface may rived here in the dilute regime for the solute, it can be generalized 54 drive fluid (or reversely particle) motion as v f ∝ ∇γLV . Now to arbitrary concentrations, see Ref. 51. extending Marangoni flows to solid-fluid interfaces is definitely As a last remark, it is interesting to note that the mechanistic not obvious, but it was recognized by Derjaguin and collabora- approach highlights an underlying fundamental symmetry in the tors 55,56 that the diffuse nature of the interface may allow the transport phenomenon. Indeed Eq. (25) introduces the osmotic fluid to “slip” over the solid surface under a concentration gradi- R L/2 pressure as the driving force on the fluid: dxcs ( ∂xU ) = ent. Diffusio-osmosis is accordingly an interfacially driven flow, L/2 − σ∆Π. Now the Smoluchowski equation for the− solute – integrated and takes its origin in the interfacial structure of the solute close over the membrane thickness L, Eq. (21) – contains the very same to the solid surface, within the first few nanometers close to the term and one may accordingly rewrite the solute flux as surface. Ds ∆Π Js = A ∆cs σ (28) − L − kBT 3.1 From electro- to diffusio- osmosis 3.1.1 From electro-osmosis... The solute flux is therefore intimately related to the osmotic pressure. As is transparent from this equation, the van ’t Hoff osmotic Let us start with the canonical example of electro-osmosis, i.e. the pressure is fully expressed, i.e. ∆Π = kBT∆cs, only when the solute fluid flow close to a solid surface generated under an applied elec- flux vanishes Js = 0 (σ = 1 and ωs = 0). Reversely for a fully per- tric field. A solution containing ions will build up a so-called elec- DsA meable membrane Js = ∆cs, and there is no osmotic pres- tric double layer (EDL) close to any charged surface: counter-ions − L sure (σ = 0 and ωs = 1). Finally this equation can be rewritten are attracted by the surface charge, while co-ions are repelled. as Js = DsA (1 σ)∆cs/L, so that the “mobility” coefficient ωs is The surface charge, say Σ, is balanced in the fluid by a density − −

Journal Name, [year], [vol.],1–43 | 7 of charge ρe = e(c+ c ), defined as the difference between the beyond the EDL and reaches its asymptotic value − − density of positive and negative ions (assuming monovalent ions here for illustrative purposes). The resulting double layer is dif- v∞ = µEOE (33) fuse and extends over a finite width, see Fig. 6. The structure of ε ζ where µEO = is the electro-osmotic mobility. In the presence the EDL was amply discussed in many textbooks and reviews, and − η of hydrodynamic slippage on the surface, the electro-osmotic mo- we refer in particular to Refs. 57–59 for further insights. As a rule bility is typically enhanced by a factor 1 + b/λD, where b is the of thumb, the extension of the EDL is typically given by the Debye slip length, see Refs. 62–64 for more details. We finally note that screening length 58,60, defined as the ζ-potential may be rewritten as a function of the electrical 1 concentration ρe (by integrating twice Eq. (31)) λD = (30) √8π` c B s 1 Z ∞ ζ = zρe(z)dz. (34) where cs is the (bulk) salt concentration in the bulk and `B = − ε 0 e2 4πεk T is the Bjerrum length (ε is the dielectric permittivity of / B From a physical point of view, electro-osmosis may be seen as a water). Typically for water at room temperature, 0 7nm and `B = . force balance between the viscous friction force at the interface the Debye length ranges between 30nm for a salt concentration and the electrostatic driving force within the EDL. The velocity 4 1 1 of 10− mol.l− to 0.3nm for a 1mol.l− salt concentration. field is expected to establish over the Debye length λD and thus the fluid friction force is typically ηv∞/λD. Now the body elec- ∼ (a) E (b) E trical force within the EDL is simply Σ E where Σ is the surface Σ × z charge. From Gauss’ electrostatic boundary condition, we have + v + x ∞ Σ = ε∂zVe z 0 εζ/λD. Altogether the force balance thus takes − | = ≈ − Σ the form - - v εζ η ∞ Σ E E (35) λD ≈ × ≈ − λD × + + + + + + λD + and this leads accordingly to the expression in Eq. (33) for the ------100 µm electro-osmotic mobility. A simple extension of this argument Σ highlights immediately the potential role of hydrodynamic slip- Fig. 6 : Electro-osmosis. (a) In the presence of a surface charge Σ, page: with a slip length b, the viscous friction force will reduce an electrical double layer forms extending typically over a distance fixed to ηv∞/(λD + b) while keeping the body force identical, so that ∼ by the Debye length λD. Under an applied electric field ~E parallel to the electro-osmotic velocity will be increased by a factor 1+b/λD. the surface, a net electric force builds up due to the unbalanced charge Altogether, the electro-osmotic flow thus takes its origin within within the electric double layer. It drives a solvent flow parallel to the the very few nanometers close to the boundary and can be there- surface, extending as a flat flow profile into the bulk. (b) Visualization of the electro-osmotic flow in a capillary, via the displacement of (neutral) fore strongly affected by molecular details: hydrodynamic slip- tagged molecules. In contrast to the parabolic Poiseuille flow, the electro- page 62, nanoscale roughness 65, contamination 66, dielectric in- osmotic flow profile takes the form of a plug flow, which barely disperses homogeneities 67, etc. This makes the underlying physics of inter- the dye. Reproduced from Ref. 61 with permission from Springer Nature, facial transport both complex and very rich. copyright 2005.

Within the EDL, there is a net charge density in the fluid, and 3.1.2 ... To diffusio-osmosis ... whenever an external electric field is applied to the fluid (parallel to the surface), this will generate a net bulk force ρeE. The Stokes equation for the fluid velocity writes accordingly in the direction While electro-osmosis corresponds to interfacially driven fluid x (parallel to the solid interface) motion under an external electric field, diffusio-osmotic motion occurs under the gradient of a solute, ∂xc∞, in the vicinity of a η∂zzv + ρeE = 0 (31) solid surface – see Fig. 7. Similarly to electro-osmosis, a key ingredient is the specific interaction of the solute with the surface, where z is the direction perpendicular to the interface. The which occurs within a diffuse layer of finite thickness. Reflect- pressure-gradient term vanishes for the shear-flow considered ing the discussion of osmosis across a model potential barrier in here. Using the Poisson equation ρe = ε∂zzVe, relating the charge − Sec. 2.4, the solute will be assumed to interact via an external density to the electric potential Ve in the fluid, one can integrate potential U (z) with the solid surface. One noticeable difference twice Eq. (31) to obtain the velocity profile to the previous membrane case though is that this potential now εE acts perpendicular to the solid surface and solute gradient (i.e. v(z) = [Ve(z = 0) Ve(z)] (32) − η − depending on z but not on x), see Fig.7. where a no-slip boundary condition was assumed here. The elec- Diffusio-osmosis with neutral solutes – We first consider the case trical potential at the interface is usually identified as the zeta of neutral solutes. The fluid velocity and solute density obey the potential Ve(z = 0) = ζ. The electro-osmotic velocity is constant coupled Stokes and Smoluchowski equations, which write in the

8 |Journal Name, [year], [vol.], 1–43 xc with the diffusio-osmotic mobility µDO given by z ∇ ∞ ∞ ∞ v 1 Z cs(x,z) 1 Z U (z) ∞ µ = z 1 dz = z exp − 1 dz. (z) DO η c η k T U 0 ∞ − 0 B − x (41) This expression is similar to Eq. (34) for the electro-osmotic mo- z bility. The effect of hydrodynamic slippage on the surface can also L x s be taken into account, along the same lines as in Refs. 69,70 and leads to an enhancement factor of the diffusio-osmotic mobility scaling as (1+b/λ), where b is the slip length and λ is the typical Fig. 7 : Diffusio-osmosis. A gradient of solute imposed far from the width of the diffuse interface. The amplification effect is expected surface induces a fluid flow. Here the solute is assumed to interact with to be massive on superhydrophobic surfaces 70 and amplification the surface via an interaction potential U (z) (an adsorbing profile on the by orders of magnitude are predicted. Interestingly for strongly figure). The solute interaction with the surface induces a force on the fluid, here towards the surface, which is higher in the more concentrated hydrophobic surfaces where the liquid-vapor interface dominates, area. This normal force converts into a parallel pressure drop which gen- the diffusio-osmotic velocity takes the physically transparent ex- erates a fluid flow from high to low concentrations (and reversely for a beff pression vDO = η ∇γLV , where beff is the effective slip length on repelling interaction). the superhydrophobic surface and γLV is the (solute concentration dependent) surface tension of the liquid-vapor interface. Similarly as in electro-osmosis, diffusio-osmosis can be interpreted in terms of a simple force balance within the diffuse layer. stationary state as: A first integration of Eq. (39) indeed shows that diffusio-osmotic 0 = ∇p + η∇2v + ( ∇U ), flow results from the balance between the viscous stress on the − − surface and an osmotic pressure gradient integrated over the dif-

0 = ∇ [ Ds∇cs + λscs ( ∇U ) + vvcs] (36) fuse layer: − · − − Z ∞ At infinity, we assume a fixed gradient ∂xc∞ along x for the solute 0 = η∂xvx wall + dz∂x [Π(x,z) Π∞(x)] (42) | 0 − concentration. These coupled equations are strongly entangled. However in Simple estimates of the various terms lead to a more qualitative the limit of a thin interfacial layer – corresponding to a range for version of this force balance as the potential U (z) which is small compared to the lateral varia- v∞ η λ ( kBT∇xc∞) (43) tions of the solute gradient –, one expects the concentration pro- λ ∼ ± × − file to relax quickly to a local equilibrium across the diffuse layer where λ is defined here as the range of the potential U , and the cs(x,z) c∞(x)exp( U (z)/kBT). ± ' − sign depends on whether the solute is attracted or depleted by the λ 2 Turning now to the fluid transport equation, the Stokes equa- surface. This leads to v∞ ( kBT∇xc∞) in full agreement ∼ ± η × − tion projected along the z direction writes simply with Eqs. (40)-(41).

0 = ∂z p + cs( ∂zU ) (37) − − (a) Microchannels (b) because the z component of fluid velocity is expected to be neg- Advection ligible for thin layers. We can integrate this pressure balance to Dye obtain + Diffusion PEG PEG nR p(x,z) p∞ = kBTcs(x,z) kBTc∞(x) (38) nL − − Nanochannel ( h = 163 nm, w = 5 μm, which can be seen as an osmotic equilibrium across the diffuse L = 150 μm) layer 68. In simple terms, the existence of a specific solute-wall imaging interaction allows the membrane to “express” the solute osmotic x z k Tc x z pressure Π( , ) = B s( , ) within the interfacial layer. However Fig. 8 : Experimental evidence for diffusio-osmosis. (a) A gradient the effects of the latter disappear in the bulk (z ∞) and there is → of polyethylene glycol polymer PEG is maintained along a nanochannel no bulk osmotic pressure gradient. thanks to lateral microchannels acting as reservoirs. The nanochannel is 160nm in thickness and is fully permeable to PEG. Under a PEG con- Now inserting the pressure from Eq. (38) into the Stokes equa- centration gradient, a diffusio-osmotic flow arises: water moves towards tion projected along x, see Eq. (36), leads to higher concentrations of PEG. The flow rate Q is measured via the concentration profile of a dye. (b) Measured diffusio-osmotic flux Q as a 2 η∂ vx ∂x[p(x,z) p∞] = 0. (39) function of the PEG concentration difference, showing a velocity propor- z − − tional to the PEG concentration difference. This behavior and the sign Following the same steps as for the electro-osmosis, one obtains of the effect are consistent with a steric exclusion of PEG on the sur- the fluid velocity along the x coordinate in the bulk fluid as faces, as predicted in Eq.(44). (a) and (b) are reproduced and adapted from Ref. 25 with permission from the American Physical Society (APS), copyright 2014. v∞ = µDO ( kBT∇xc∞) (40) × −

Journal Name, [year], [vol.],1–43 | 9 Diffusio-osmosis is definitely an osmotic flow, e.g. a flow driven the gradient of the salt concentration (if no current exists in kBT by an osmotic pressure gradient located within the diffuse layer. the bulk). This takes the form Ediff = e δ∇logc∞ with δ = However the direction of the diffusio-osmotic flow can be along (D+ D )/(D+ + D ) and adds a supplementary electro-osmotic − − − or against the gradient of the solute, in strong contrast to bare os- contribution to the diffusio-osmotic velocity as v = εζ E . diff − η × diff mosis which induces a flow towards the highest solute concentra- Accordingly this leads to a supplementary contribution to the motion. That is highlighted in the expression of the diffusio-osmotic bility as: mobility, Eq. (41), which can be positive or negative depending diff εζ kBT DDO = δ (47) on the attractive or repulsive nature of the interaction potential − η × e U (z). As a rule of thumb, the sign of the mobility will be domi- R ∞ nantly determined by the adsorption Γ = 0 dz(c(x,z)/c∞(x) 1). (a) (b) − xc If there is a surface excess (Γ > 0 or U (z) < 0), the solvent flow ∇ ∞ goes towards the low concentrated area (µDO > 0). That may ap- - v + + ∞ pear as surprising because it amounts to concentrating even more + - the already concentrated solution; we shall discuss this appar- - - ent paradox in Sec. 3.2.2. Reversely a surface depletion resulting + + z + + from a repulsion of the solute from the wall (Γ < 0 or U ( ) > 0) λD + + + + reverses the direction of the solvent flow towards the high con- + ------centrated zone (µDO < 0). An interesting limiting case for this Σ behavior is exemplified by a solute interacting with the wall via steric effect, i.e. hard-core excluded volumes. For a solute particle Fig. 9 : Diffusio-osmosis with charged electrolytes. (a) Geometry; an with radius R, the mobility in Eq. (39) reduces to electrolyte concentration difference is imposed far from the charged interface. The electrical interaction with the surface induces an (attractive) 2 R electrostatic force – sketched with arrows– on the fluid which is larger µsteric = . (44) DO − 2η where the salt concentration is larger, hereby inducing a net flow towards the low salinity region. (b) Measurement of the diffusio-osmotic flux as This behavior was measured in particular in Ref. 25 for the a function of the difference of the logarithm of the salt concentration be- diffusio-osmotic flow under a neutral polymer concentration gra- tween two reservoirs, in a similar way as in Fig. 8. Note the reversal of the dient, see Fig. 8 for an illustration. A final remark is that this sign as compared to Fig. 8-b. Reproduced from Ref. 25 with permission from the APS, copyright 2014 simple rule for the correlation between adsorption and the sign of diffusio-osmosis is not exact and may fail for more complex interactions between the solute and the wall, for instance with an An important remark is that for electrolytes the velocity is pro- oscillatory spatial dependence of the concentration profile due to portional to the gradient of the logarithm of salt concentration, layering. The sign of µDO may then be expected to differ from the in contrast to solutes where it is basically linear in the gradient, sign of the adsorption Γ. In this case, no obvious conclusion can see Eq. (40). We will refer to this dependence as “log-sensing” by be made for the direction of the diffusio-osmotic velocity and a analogy to behaviors occurring for the chemotaxis of biological full calculation has to be made, see for example Ref. 26. entities (e.g. bacteria). Such a dependence may be understood on the basis of the simple scaling argument based on the force bal- Diffusio-osmosis with electrolytes – We now discuss specifically ance above, see Eq. (43). Indeed the thickness of the diffuse layer 2 the case of diffusio-osmosis under salinity gradients. Here, as λD is now given by the Debye length, and v∞ ( kBT∇xc∞). for electro-osmosis, the diffuse layer corresponds to the electric ' η × − Since the Debye length depends on the salt concentration as double layer created close to a charged surface, see Fig. 9. The λ 2 = (8πl c ) 1, one obtains : derivation follows similar steps as above, from Eqs. (36) to (41), D B ∞ − except that one has to take into account the spatial distribution of kBT v∞ ∇x logc∞. (48) both the counter- and co- ions in the EDL that follow a Poisson- ≈ − 8πηlB Boltzmann distribution, see Ref. 71. In this case the diffusio- which is qualitatively similar to the exact results in Eq. (46) and osmotic velocity is shown to take the form predicts log-sensing for diffusio-osmosis with electrolytes.

v∞ = DDO( ∇logc∞) (45) This behavior is confirmed by experimental investigations of − water flows under salinity gradients in nanofluidic circuits 25, see where we introduced a mobility DDO which has now the units of Fig. 9. Diffusio-osmotic flow of water under salinity gradients was a diffusion coefficient. It takes the expression 54 also evidenced across carbon nanotube membranes 72, confirm- ing further that diffusio-osmosis was acting against bare osmosis. kBT 2 Φ0 DDO = log cosh (46) In an alternative configuration, diffusio-osmosis was also shown 2πη`B × 4 to induce very large ionic currents under salinity gradients 73–75. where Φ0 = eV0/kBT is the dimensionless surface potential V0 We will come back to such cross effects associated with diffusio- (usually identified with the zeta potential). Note that for an osmosis in Sec. 3.2.1, as well as in the section dedicated to blue electrolyte with unequal diffusion coefficients for the anions and energy harvesting, Sec. 6.3. In a very different field, diffusio- cations (D+ = D ), a diffusion electric field builds up under osmotic flows were also shown to strongly impact and shape the 6 −

10 |Journal Name, [year], [vol.], 1–43 reactive fluid flows occurring in the solid Earth 76. Log-sensing the fluid rewrites has also many counter-intuitive consequences and a variety of B applications 77,78, which we will discuss more specifically in the fT = ∑ ∆ci fi(z) = ∑ ∆ci ( ∇µel,i) (50) i=1,n × i=1,n × − context of diffusio-phoresis in Sec. 5. B where the sum runs over n ion species and ∆ci = ci(x,z) c (x) is Solvo-osmosis and diffusio-osmosis with mixtures – Up to now we − i considered merely dilute solute solutions, but all previous results the excess ion concentration in the boundary layer, as compared can be generalized to mixtures of liquids with any molar fraction to the bulk. This driving force will generate a flow according to 2 of its constituents. The key ingredient remains that the two con- the Stokes equation η∂z vx + fT = 0 and following the same steps stituents interact differently with the solid substrate. As shown in as above, one obtains the far field slip velocity as Ref. 51, the diffusio-osmotic velocity now takes the expression Z ∞ 1 B B v∞ = ∑ ∇µel,i dzz∆ci(z) ∑Mi( ∇µel,i), (51) − i=1,n η 0 ≡ i − v∞ = µDO( ∇xΠ[X∞(x)]), (49) − where the mobility M takes the expression M 1 R ∞ dzz∆c z . where Π is the generalized osmotic pressure defined in Eq. (9), i i = η 0 i( ) For symmetric and monovalent ions, these mobilities can be ex- calculated for the molar fraction X∞, hence generalizing the actly calculated using Poisson-Boltzmann framework, leading to expression in Eq. (40). The diffusio-osmotic mobility µDO is 1 R ∞ cs(x,z0) still given by the initial expression µDO = 0 dz0 z0 1 . ε ζ kBT Φ0 η c∞(x) − M = + log cosh2 (52) However, for a solute-substrate interaction potential U , the con- ± eη ∓ 2 e × 4 centration profile cs(x,z) is now implicitly related to the value with Φ0 = eV0/kBT the dimensionless surface potential and here in the bulk c∞(x) via the local equilibrium condition µ[cs(x,z)] + ζ V the zeta potential. z c x 0 U ( ) µ[ ∞( )]. ≡ B ' Under a constant electric field ∇µ = eE0 and the electro- Diffusio-osmosis with ethanol-water mixtures was investigated ±1 ∓ osmotic mobility is predicted as µEO = e (M+ M ), in full agree- − − recently in Ref. 26. But the majority of existing experimen- ment with the previous result in Eq. (33)-(34). Under an imposed B tal investigations merely explored the reverse configuration of ionic strength gradient in the bulk, then ∇µ = kBT∇logc are ± ± phoretic transport of particles under gradients of liquid com- identical and MDO = M+ + M DDO, again in full agreement 21,79 − ≡ position, denoted as “solvo-phoresis” . Interestingly in Ref. with the previous result in Eq. (46). 21, the phoretic transport of colloidal (polystyrene) particles in ethanol-water mixtures resulted in a “log-sensing” behavior of the 3.2 Transport matrix and symmetry considerations particle diffusio-phoretic velocity, obeying V = DSP∇logX, with 3.2.1 Transport matrix and cross fluxes here X the ethanol mole fraction. As introduced in Sec.2.3, the framework of irreversible processes 3.1.3 ... And electro-chemical equivalence. allows one to write a linear relation between thermodynamic forces and fluxes 30. Adding the electric forces to the set of forces, In the case of electrolytes, the two previous transport phenomena, one may generalize the results in Eq. (14) to obtain linear trans- electro- and diffusio- osmosis, are fundamentally intertwined. In- port equations now relating the solvent flux Q, excess solute flux deed, from the thermodynamic point of view, the chemical poten- Js csQ and electric current Ie to the pressure gradient ∇p, − − tial and the electric potential contributions merge into the elec- chemical potential gradient ∇µ and the applied electric field − trochemical potential: µ = µ + qV (with q the ion charge and ∇Ve, and summarized as el − V the electric potential). There is accordingly a deep analogy     Q ∇p when driving the system under gradients of chemical potential −  Js csQ  = L  ∇µ , (53) (diffusio-osmosis) or driving under gradients of electric potential  −  ×  −  Ie ∇Ve (electro-osmosis). An illuminating discussion on this point and − the corresponding force balance is provided by T. Squires in Ref. Due to Onsager principle, this matrix is symetric and positive 59,80 and we reproduce the essentials of the argumentation here. definite 30. Each term of this matrix corresponds to a specific Let us consider in full generality that a gradient of the electro- transport phenomenon. Diagonal terms are associated respec- chemical potential is applied in the bulk far from the boundary, tively with permeability (characterizing solvent flux under a pres- B sure drop), diffusion (characterizing solute flux under an applied ∇µel,i (along the direction of the solid surface, say x); the index i runs over the various ion species in the solution. As we dis- solute gradient) and electrical conductance (characterizing ionic cussed above for both electro- and diffusio- osmosis, this will gen- current under an applied electric field). The off-diagonal terms erate net thermodynamic forces on individual ion specie i, which correspond to cross effects. We detail below the cross effects that may be written as fi(x,z) = ∇µ (x,z). A key remark is that are all recapitulated in Fig. 10. − el,i the electrochemical potential is approximately constant across the In the first row of the matrix, electro-osmosis and diffusio- B EDL, i.e. µel,i(x,z) µel,i(x), so that the individual force rewrites osmosis – explored so far – correspond to the terms relating the B ' fi(z) ∇µ . The interfacial motion results from the forces in solvent flux Q to a chemical gradient ∇µ and an electric field ' − el,i − excess to the bulk, so that the corresponding total force acting on ∇Ve respectively. A key consequence of the symmetry of the ma- −

Journal Name, [year], [vol.],1–43 | 11 v , one indeed recovers Eq. (56). However the prediction of Diffusio- Electro- DO Q Permeability p osmotic flow osmotic flow −∇ Eq. (56) reports a more complex dependence, since the linear de- Excess flux Excess flux pendence in Σ is only valid for large enough Σ, while for low Σ, under Diffusion under 3 Js csQ = µ one finds that Kosm ∝ Σ , i.e. vanishingly small. Such osmotically − pressure electric field × −∇ driven currents have been measured experimentally in various Diffusio- Streaming Electric I osmotic V systems, nanochannels, single nanotubes, single nanopores – see () e () ()ion current conductance () () e () ion current −∇ Refs. 73,74,82,83 – to cite a few. This effect finds important applications in the context of blue energy harvesting 75, that we will Fig. 10 : Transport Matrix. Explicit transport matrix L as presented in Eq. (53), with colors indicating symmetric terms. explore in detail in Sec. 6.3. 3.2.2 Entropy production with diffusio-osmosis We pointed out above that the sign of the diffusio-osmostic mo- trix is that the same mobilities characterize symmetric transport bility, µ , can be either positive or negative, so that the corre- phenomena. For example consider the first column of the matrix DO sponding flux can be along or against the concentration gradient. L, one finds that the electro-osmotic mobility and diffusio-osmotic A negative µ may appear at first sight striking since the direc- mobility also describe respectively the electric current and excess DO tion of the solvent flow corresponds to that of an increase in salt solute flux generated under a pressure drop, as concentration, thus leading to an apparent violation of the second principle. This is however not the case, as it can be verified Ie = A µEO ( ∇p) × − from a calculation taking into account all relevant fluxes. To high- Js csQ = A µDO ( ∇p) (54) light this situation, let us consider a membrane separating two − × − reservoirs with fixed volumes; the concentration on the left/right where is the channel cross section. The first corresponds to the ∆cs(t) A reservoir is cs(t) = c0 . The pore size is assumed here to be ∓ 2 so-called streaming current and takes its origin in the motion of larger than the solute diameter so that the membrane is perme- mobile ions in the EDL which are carried by the pressure-driven able to the solute and there is no bare osmotic pressure. A salinity flow; the pressure-driven excess solute flux has a similar physical gradient however generates a diffusio-osmotic flow on the pore origin. surface. Based on the transport matrix formulation, Eq. (14), one Streaming currents are commonly measured in experi- may write the solvent and (excess) solute fluxes as a function of 60,62 ments , even down to single carbon and boron-nitride nan- the solute concentration and pressure gradients according to: otubes 73. To our knowledge, no experimental measurement of pressure-driven excess solute flux has been performed up to now. Ap Q = [κ( ∆p) + µDO( kBT∆cs)] However this is not the case in molecular dynamics simulations L − − where it is far easier to measure the diffusio-osmotic mobility via Ap 26,69 Js c0Q = [µDO c0( ∆p) + λs( kBT∆cs)] (57) the pressure-driven excess flux – see details in Sec. 3.5. − L − − Now, the transport matrix suggests that an electric current can with Ap the total (open) pore area of the membrane, L its thick- be generated under an osmotic gradient, which we term here the ness and λs = Ds/kBT the diffusive mobility of the solute across diffusio-osmotic ion current, following the membrane, defined in terms of the solute diffusion coefficient; κ is defined in terms of the permeance as κ /L IDO = Kosm ( ∇logc∞). (55) Lhyd Ap × − ≡ (note that κ = κhyd/η where κhyd is the permeability introduced Let us consider a channel in the form of a slit of width w and above). The second principle imposes that the transport matrix height h (with w h to simplify). Using Poisson-Boltzmann to in Eqs. (14) & (57) should be definite positive. Accordingly, the 2 describe the EDL, one can calculate the corresponding osmotic determinant det(L) ∝ κλs µ c0 must be strictly positive. − DO electric current 64,73,81 and the mobility takes the form On the other hand, since the volume is fixed, the flux vanishes, Q = 0, and the solute flux writes 1 kBT sinh− χ Kosm = α ( Σ) 1 (56) × − 2πη `B − χ Ap 1 h 2 i Js = λs κ µ c0 ( kBT∆cs) (58) L κ − DO − where α 2w is the perimeter of the channel cross section. In ' Φ0 We find that the term in brackets is proportional to the determi- this expression we introduced χ = sinh | 2 | with Φ0 = eV0/kBT nant det( ), and therefore is constrained by the second princi- the dimensionless surface potential V0. In the Poisson-Boltzmann L framework χ is related to the surface charge Σ according to ple to be positive. Accordingly, whatever the sign of the diffusio- osmotic mobility µDO and the corresponding diffusio-osmotic sol- χ = 2π`BλD Σ /e with λD the Debye length, so that χ ∝ Σ . This | | | | formula can be extended to take slippage on the surface into ac- vent flux, the total solute flux will go down the solute gradient, count, as well as mobile surface charges 64. More precisely the as expected from the second principle. diffusio-osmotic ion current takes its origin in the motion of ions in the EDL which are carried by the diffusio-osmotic flow. As a 3.3 The peculiarity of diffusio-osmosis across an orifice simple estimate we may write that IDO ( Σ) vDO, where vDO In the previous sections, we implicitly considered (diffusio- ≈ − × is the diffusio-osmotic velocity: using the expression Eq. (40) for osmotic) transport across long channels, so that fluid flow is

12 |Journal Name, [year], [vol.], 1–43 translationally invariant along the channel’s length. However ξ curves are respectively oblate spheroids and hyperboloids of transport across thin membrane pores 84–86 raises the question of revolution). This expression involves a complex spatial average the specificity of these geometries in which the channel length of the Boltzmann weight e U /kBT 1, which should be compared − − L may decrease down to molecular lengths, in particular with to the corresponding simple expression in Eq. (41) for the planar the advent of 2D materials such as graphene, h-BN and MoS2 case. 74,87,88 as membranes for fluidic transport . For example, recent The above result can be simplified for certain functional forms measurements across nanopores in MoS membranes reported 2 of the potential U . For example, assuming that the interaction huge diffusio-osmotic ion currents under salinity gradients 74. In potential U depends only on variable ξ allows the mobility to another experiment, gradients of salts were shown to strongly be rewritten in terms of the two-dimensional interaction within increase the capture rate of DNA molecules across solid-state pore 1 R a p 2 2 U (ρ)/kBT the pore only as µDO = 2 0 dρρ a ρ e− 1 nanopores 84. πa η − − with ρ the axi-symmetric distance to the center of the pore 93. For long channels the driving force for fluid transport, e.g. the This expression for the mobility can be recovered thanks to the gradient of the chemical potential, is expected to scale as its in- symmetry of the transport matrix Eq. (53). Indeed µpore can also verse length, ∇µ = ∆µ/L. This would suggest that the driving DO be calculated in terms of the excess solute flux under a pressure force diverges as 1/L in the limit of nanopores where L 0. → driven flow: in this case the velocity profile was shown to be However entrance effects level off this diverging behavior to a semicircular (and not parabolic) 94–96, see Fig. 11, and the excess value typically fixed by the lateral size of the pore, say a its ra- solute flux conveyed by the circular flow reduces to the above dius – see Fig. 11. As a rule of thumb, one may expect that expression. ∇µ ∆µ/a (see for example Ref. 89 for the conductance of ion ≈ channels). However the flow in and out of the pore is expected to The complexity associated with diffusio-osmosis across an ori- be strongly disturbed, as shown for example for electro-osmosis fice is also highlighted by the predicted dependence of the mo- across nanopores in thin membranes 90–92. Similar effects are bility on the pore size a and the range λ of the interaction U . accordingly expected to apply to diffusio-osmotic transport. Let us focus the discussion for the thin diffuse layer case, where λ a (we refer to Ref. 93 for a full discusson). As a reference, (a) velocity (mm/s) (b) the diffusio-osmotic mobility across a long channel with length λ 2 0 50 100 150 L was shown previoulsy to scale as µDO , see e.g. Eq.(43). ∼ ηL ρ However, for an orifice in a thin membrane, Rankin et al. showed a pore on the basis of Eq.(59) that the mobility exhibits a variety of non- pore γ 1 γ trivial scalings, with µ λ a − , and an exponent γ that de- channel v (ρ) DO ∼ ⊥ pends on the details of the interaction potential U . For example, for the potential discussed above, which assumes a dependence as U (ξ), one finds γ = 3/2; but for a potential depending on the distance to the membrane or to the edge of the pore, then Fig. 11 : Peculiarity of pressure flow across an orifice. (a) Simulated γ = 2 93. In the latter case, γ = 2, the diffusio-osmotic mobility flow velocity and streamlines across a pore with, say, radius a = 10nm scales as µpore λ 2/a, which corresponds to the long channel re- and thickness L = a/10 under a pressure drop ∆p = 1bar. The scale bar DO ∼ is 10nm. The streamlines are spaced equally in magnitude at the center sult with the length L replaced by a. But for other values of the of the pore. (b) Simulated flow velocity and streamlines across a channel exponent γ this simple rule of thumb does not apply, making the with same radius a and thickness L = 10a under the same pressure drop. diffusio-osmotic transport across the orifice quite peculiar. The scales (velocity and geometry) are the same as for (a). For readabil- ity the whole channel length is not plotted. (Center) Normalized velocity As a last comment, it is possible to extend qualitatively these profile (perpendicular to the membrane) at the center of the membrane, results to electrolyte solutions, by assuming that the potential comparing the channel and pore cases. range λ identifies with the Debye length. This suggests an anomalous salinity dependence for the diffusio-osmotic mobility The diffusio-osmotic flow across a nanopore with vanishing 1 γ/2 DDO = vDO/[ kBT∆logcs] ∝ cs− , in contrast to long channels thickness was recently calculated analytically by Rankin et al. 93. − 0 where DDO ∝ cs . The nanopore geometry may thus depart from The calculation is best performed in oblate-spheroidal coordi- the log-sensing behavior of diffusio-osmotic transport under salin- nates, in line with a similar calculation for electro-osmosis in Ref. ity gradients. These results remain however to be fully assessed 90. The averaged diffusio-osmotic velocity vDO across the pore, experimentally. 2 which is defined in terms of the diffusio-osmotic flux Q = πa vDO, is proportional to ∆cs the difference (and not the gradient as in Eq. (40)) of the solute concentration between the two sides of 3.4 Alternative interfacial transport: thermo-osmosis pore the membrane: vDO = µ ( kBT∆cs). The general expression DO − Extending on electro- and diffusio- osmosis, thermo-osmosis cor- for the mobility derived in Ref. 93 takes the form responds to fluid motion under gradients of temperature; see Fig. 12-a. Such effects were reported as early as in the 1900s 97,98. Z 1 Z ∞ U /kBT pore 2a 2 e− 1 µDO = dξ ξ dν − (59) Thermo-osmosis was first rationalized in terms of thermodynamic − π2η 0 0 1 + ν2 forces by Derjaguin et al. 54,99,100. Similarly as for diffusio- where (ν,ξ) are the oblate spheroidal coordinates (iso-ν and iso- osmosis in Eq. (41) and electro-osmosis in Eq. (34), the net ve-

Journal Name, [year], [vol.],1–43 | 13 locity generated far from the surface is predicted as 54 Fig. 18-a). This phenomenon was harvested to manipulate colloids and build structures 107,111–113 with advanced applications 2 Z ∞ v = − zδh(z)dz ∇logT (60) in microfluidics 101 or towards DNA detection 114. Among other ∞ η ∞ 0 phenomena, it was shown that couplings between thermo- and where T∞ is the temperature far from the surface and δh(z) is diffusio- phoretic drivings allow to finely manipulate colloidal the excess specific enthalpy in the interfacial layer as compared structures 115. Also, thermophoresis of molecules can provide de- to the bulk liquid. If the solid surface is e.g. hydrophilic, then tailed information about particles and molecules (size, charge and δh(z) . 0 and the flow of water is directed toward higher tem- hydration shell) and this provide very efficient analytical tools to peratures, see Fig. 12 and Refs. 101,102. An interpretation of probe protein in biological liquids 116,117. In a different context, thermo-osmosis (and -phoresis) in terms of interfacial surface ten- applications of thermo-osmosis were suggested for the recovery sion modification, and therefore Marangoni-like flow generation, of water from organic waste-water 118, as well as for energy har- has also been suggested 103 and formalized 104,105. The transport vesting from thermal differences (and waste heat). 110,119,120 of fluids or particles under thermal forces led to strong debates between the interfacial approach discussed above and an “ener- 3.5 Numerical simulations of (diffusio-)osmotic transport: 106–108 getic” approach , which attempts to write the net driving methodologies and results force acting on a particle as the gradient of a thermodynamic Molecular simulations have now become a highly efficient tool quantity 106. The resulting Soret coefficient – defined as the ratio to explore the fundamental properties of fluids and materials. between the thermophoretic mobility and particle diffusion coef- Molecular dynamics simulate the many-body dynamics of par- ficient – highlights a different dependence on the particle size as ticles and molecules, either at equilibrium or far from equilib- compared to the interfacial framework discussed above. Although rium, submitted to various thermodynamic forces. They provide attractive, the energetic approach was then extensively criticized detailed information on the molecular processes at play. In the 107,108. present context of studying osmotic forces and related fluxes, this (a) T + ∆T T (b) represents a key opportunity to understand the fundamental and subtle origins underlying interfacial transport and how these can v be affected by the microscopic details of the interface. ∞ Simulating electro-osmosis is relatively straightforward in the Heat spent Heat released by desorption by adsorption sense that the effect of the electric field converts directly into an electric force acting on the suspended ions. This has led to numerous molecular dynamics studies of electro-osmosis, as well as of streaming currents, allowing to decipher a wealth of phenomena Hydrophilic surface associated with transport within the electric double layer 62,121. Now, simulating diffusio- and thermo- osmosis is by far more difficult and subtle. Indeed such transport occurs under thermo- Fig. 12 : Thermo-osmosis near an interface (a) Geometry; a temperature gradient is imposed far from a hydrophilic surface. A thermal dynamic forces associated with the gradient of concentration or flux from the hot to the cold region is therefore installed. The inter- temperature, and these can not obviously be represented in terms action of water with the surface induces a force (light red arrows) that of mechanical forces acting on the simulated particles. We discuss varies along the surface due to the thermal gradient, inducing a net flow. in this section recent developments in the numerical methodolo- (b) Thermo-osmotic flow measured on two different surfaces (going to- gies allowing to perform simulations of osmotic transport. wards the higher temperatures) as a function of the distance to the heat source. Reproduced from Ref. 101 with permission from the APS, copy- For bare osmosis, direct simulations can be performed using right 2016. two explicit reservoirs with difference of solute concentration. For example such implementation was used by Kalra et al. in As for electro- and diffusio- osmosis, the details of the interfa- the study of osmosis across carbon nanotubes 122. This configura- cial dynamics, for example slippage at the interface, is expected tion has the drawback that osmosis occurs in the transient regime to strongly affect thermo-osmotic flows. This has been evidenced since the reservoirs empty/fill during the osmotic process and this for example in molecular dynamics where a huge enhancement limits statistics. Osmosis was later rationalized in more simple of thermo-osmosis was measured with slip 109,110, although the terms by simplifying the explicit membrane description to reduce exact dependence of thermo-osmosis on the interfacial proper- it to a confining potential acting on the solute only 123–125. This ties was measured to be substantially complex. 110 We refer to the is the numerical pendant to the mechanical views of osmosis de- Refs. 102,107 for more in-depth discussion on thermo-osmosis scribed in Sec. 2.4. and -phoresis. The numerical implementation of diffusio-osmosis in molecular Lately thermo-osmosis has gained growing attention in terms dynamics is far more complex since one should be able to repre- of applications and we briefly comment here on this aspect. sent the chemical gradient in terms of a microscopic force acting Many applications of the phenomenon are done in the context on the particles. Various methods to investigate diffusio-osmotic of thermo-phoresis, or displacement of colloidal particles un- transport were proposed in the recent literature and we discuss der thermal gradients (in a similar way to diffusio-phoresis, see them now.

14 |Journal Name, [year], [vol.], 1–43 Using symmetry relations to infer transport coefﬁcients – ulations. This approach was followed in Refs 125,129,130 and It turns out that it is far easier to calculate the diffusio-osmotic the resulting mobilities were successfully compared with results mobility by exploiting the symmetry of the transport matrix. Re- of NEMD simulations, as discussed below. calling the general relation between ﬂuxes and forces, ! ! ! (a)Pressure gradient p (b) Chemical potential gradient µ Q L11 L12 ∆p −∇ −∇ = − . (61) Js csQ L21 L22 × kBT∆logcs

− − the Onsager symmetry for the transport matrix implies that L21 = L12. Accordingly, calculating the diffusio-osmotic mobil- Force free Force bulk region ity as a water flux under a concentration gradient, here L12 = Q/( kBT∆logcs), is therefore equivalent to calculating the excess − solute flux under a pressure gradient, here L21 = (Js csQ)/( ∆p) − − – see Fig. 13-a. The latter is far easier to implement numerically in non-equilibrium molecular dynamics (NEMD) since it requires only to generate a pressure-driven flow and measure the solvent solute solvent solute U wall U wall U wall U wall intergrated solute flux (or locally the velocity and solute concentration profile). This can be performed with periodic boundary Fig. 13 : Non-equilibrium molecular dynamics of osmotic interfacial conditions along the flow, so that the resulting diffusio-osmotic transport. Inspired from Ref. 125. (a) Excess flux under pressure gradi- mobility is indeed characteristic of the liquid-solid interface un- ent. The pressure gradient is obtained by applying a force acting on each particle and the solute flux in excess to the bulk Js csQ is measured. (b) der scrunity, and does not depend on e.g. entrance effects into − NEMD method to simulate interfacial transport under chemical potential the pore. This methodology was successfully applied to quan- gradients. The chemical potential gradient is modeled as a forward force tify the diffusio-osmotic mobility on a variety of interfaces, in- per solute particle (red) and a properly defined counter force per solvent cluding superhydrophobic surfaces, graphene, and with various particle (blue) such that the total force on the fluid in the bulk is zero. The liquids 26,69,70,110. We discuss below some results of the simula- local diffusio-osmotic velocity profile is directly measured. tions.

Equilibrium fluctuations for linear response coefficients – Non-equilibrium molecular dynamics and mechanical representa- Transport coefficients may also be inferred from equilibrium fluc- tion of chemical gradients– tuations by making use of Green-Kubo (GK) relations for the vari- While the equilibrium approach provides proper foundations to ous mobilities. The transport coefficients introduced in the trans- calculate diffusio-osmosis, non-equilibrium simulations proves port matrix L can indeed be written in terms of a time-correlation usually more practical to calculate transport coefficients, e.g. in function of the fluctuating fluxes Qi at thermal equilibrium. Such terms of statistics. However, as we emphasized above, this re- formal relations are obtained thanks to linear-response theory quires to build a proper numerical scheme to implement a me- and the fluctuation-dissipation theorem 126–128. They provide chanical equivalent of the chemical potential gradient. One in- generic expressions for the non-equilibrium mobilities in terms teresting route was suggested by Yoshida et al. in Ref. 129 and of equilibrium correlation functions in the form then applied to electro- and diffusio- osmotic transport of electrolytes: the authors ran different simulations where forces f j V Z ∞ Li j = dt Qi(t)Q j(0) (62) are applied separately to each individual specie, here {solvent, k T 0 h i B anions, cations}, allowing to calculate the corresponding individ- where V is the system volume and Qi are the fluxes under ual fluxes Qi and deduce the mobilities for the individual species { } scrutiny. The symmetry of the transport matrix originates in the Mi, j = Qi/ f j; the electro- and diffusio- osmotic mobilities are then time-symmetry of the underlying microscopic dynamics 20. The calculated by proper linear combinations of the mobilities of indi- simplest route to obtain the Green-Kubo formula for the diffusio- vidual species, in order to deduce the electro- and diffusio- osmo- osmotic mobility is to consider the solute excess flux generated sis. This approach echoes directly the discussion in Sec.3.1.3, in under a pressure drop since the latter is equivalent to a body which the electro-osmotic and diffusio-osmotic mobilities are de- force applied to all system particles. The linear-response formal- duced from the individual ion mobilities, defined above as M 129. ± ism then immediately leads to 125 It is however relevant to develop numerical methods to simulate explicitly the diffusio-osmotic flows. Such a numerical V Z ∞ L21 = L21 = (Js c∞Q)(t)Q(0) dt. (63) scheme was recently proposed in Ref. 125, in which a proper kBT 0 h − i mechanical set of driving forces is applied to the system to mim- In the case of a channel with length L and cross area A , one ick the chemical potential gradient of the solute. To do so, the A has L21 = L21 µDO. We refer to Refs 125,129 for detailed scheme applies differential forces on the solute and on the sol- ≡ L derivations of these GK equations. vent, see Fig. 13-b: (i) an external force Fµ on each solute particle B B B These GK formula allow to calculate numerically the diffusio- in the whole system; (ii) a counter force [Ns /(N Ns )] Fµ , B − B − × osmotic mobility, as well as any off-diagonal terms of Eq. (53), by acting on each solvent particle. Here Ns and N are respectively estimating the correlation functions in Eq.(63) in equilbrium sim- the number of solute particles and the total number of particles in

Journal Name, [year], [vol.],1–43 | 15 a properly defined bulk region (“sufficiently” far from the surface). considered as surprising in view of the strong velocity gradients The counter force is set to ensure a force free balance in the bulk occurring on length scales in the range of a few molecular size. volume. Most important, it can be verified that applying linear- However the Stokes equation is known to be surprisingly robust response theory to the system with this set of forces allows one down to molecular lengthscales 62 and this explains its success in to show that the resulting diffusio-osmotic mobility does identify predicting diffusio-osmotic flows. with the GK relation in Eq.(63): this therefore fully validates the Finally, the continuum framework allows one to relate the theoretical foundations of the proposed numerical scheme. The diffusio-osmotic mobility to the concentration profile, and more corresponding effective chemical potential is then related to the particularly to its first spatial moment, see Eq.(41). It is accord- applied external force Fµ via ingly tempting to relate – as for Marangoni effects – the amplitude of diffusio-osmotic transport to the adsorbed quantity, de- NB R ∇xµ = Fµ . (64) fined as Γ = dz(cs(z) c∞). The latter is directly connected to − NB NB − − s the surface tension via the Gibbs-Duhem relation. The adsorbed quantity Γ provides in most cases a good estimate for the diffusio- This approach leads as expected to a velocity profile exhibit- osmotic mobility and its sign. However – as mentioned earlier – ing a strong gradient within the interfacial layer, and then a plug for complex concentration profiles the relation was found to be flow far from the surface. The deduced diffusio-osmotic mobil- more complex than this simple rule of thumb (for example for ity obtained from the NEMD scheme was checked to be identical water-ethanol mixture at interfaces 26). to both the equilibrium GK results and those obtained from the 125 excess flux under pressure-driven flow introduced above . 4 Osmosis beyond van ’t Hoff Some difficulties with the microscopic stress tensor– 4.1 Advanced osmosis and nanofluidics The continuum approach, as described above, allows one to pre- The previous section highlighted molecular insights into osmotic dict diffusio-osmotic transport in terms of a surface pressure gra- phenomena, unveiling the underlying driving forces at play. How- dient. In a different approach, it is accordingly tempting to obtain ever such perspectives also suggest possible extensions to obtain the diffusio-osmotic flow by a direct numerical calculation of the more advanced osmotic transport beyond the linear framework local microscopic pressure in the fluid. However, as was demon- presented before. In this section we discuss osmotic transport strated by Frenkel and collaborators in a series of papers 131–134, across channels with more complex geometries involving sym- a major difficulty in this approach is that there is no unique ex- metry breaking, or active parts. Our objective in this section is pression for the local microscopic pressure tensor (e.g. in terms of to show that it is possible to extend simple osmosis beyond the the position and velocities of individual particles and the micro- van ’t Hoff paradigm and design advanced functionalities result- scopic forces acting on them). Accordingly various microscopic ing in non-linear and active transport. definitions of the pressure tensor lead to different numerical re- In this context it is interesting to observe that in biological sults. Such difficulty was evidenced for diffusio-osmotic flows 132, species (bacteria, archaea, fungi, ...) many membrane chan- but also for thermo-osmotic flows 133,134. nels do achieve advanced functionalities in order to regulate os- Some results of simulations– mosis: for example rectified osmosis 135 –e.g. an osmotic flow Simulations have allowed to gain much insights into diffusio- with a non-linear dependence on the concentration gradient–, osmotic transport. Various fluids, e.g. Lennard-Jones fluids, but or gated osmosis to prevent lysis and survive osmotic shocks in also with electrolytes and water-ethanol mixtures, and various mechanosensitive channels 3 (with diffusio-osmosis identified as interfaces were considered, hydrophilic or hydrophobic surfaces, a potential mechanism for the gating mechanism in deformable graphene, superhydrophobic surfaces, etc. Among highlighted ef- structures 136). These few examples highlight the possibility of fects one may quote the impact of hydrodynamic slippage of the going far beyond the van ’t Hoff paradigm, thanks to properly de- fluid at the surface, which does boost considerably the diffusio- signed (active) nanochannels. osmotic mobility on hydrophobic 69 and graphene surfaces 125, We believe that the advent of nanofluidics has a key role to and even more on super-hydrophobic surfaces 70. The enhance- play in this regard, in order to identify new types of behaviors ment of the diffusio-osmotic mobility scales typically like the ratio which could be scaled-up to macroscopic membranes. The new between the (effective) slip length and the interfacial length, as opportunities brought by nanofluidics in terms of the variety of mentionned in the previous sections. nanoscale geometries and materials, combined with state-of-the- Simulations also give some insights on the local diffusio- art experimental instrumentation, allows one to fabricate and in- osmotic velocity profile and its relation to the concentration pro- vestigate fundamentally the transport in ever smaller channels, files. Within the continuum framework discussed in the previous with ever more complex and rich behaviors. Carbon nanotubes, section, the velocity profile is obtained simply by integration of down to nanometric sizes 73,138–140 can now be manipulated and the Stokes equation of motion in Eq.(39), with the pressure ex- inserted in devices were water is flown through – see Fig. 14- pression given in Eq.(38). Simulations actually show usually a a. Single nanopores can be carved or etched in membranes that very good agreement between the continuum prediction and the are only an atomic layer thick 74 and may be accordingly func- velocity profiles measured in the NEMD simulations, giving strong tionalized 141, see Fig. 14-b. It is also now possible to fabricate support to the continuum description. Such an agreement may be angström˙ scale slits using graphene sheets as spacers, reaching

16 |Journal Name, [year], [vol.], 1–43 (a) (b) (c) channels with an asymmetric design, e.g. an asymmetric surface charge or an asymmetric geometry. Such behavior is expected to occur in the regime where the Dukhin number is of order one and asymmetric along the channel 154: the Dukhin number is defined here as Du = Σ/csh, where Σ is the surface charge density, cs the bulk salt concentration and h a characteristic channel dimension. It quantifies the importance of surface versus bulk electric con- duction. As such ionic diodes may find interesting applications to boost osmotic power generation under salinity gradients, see Ref. 75 and Sec. 6.3. Now, coming back to osmosis, the asymmetry of the design may 1 μ m be expected to yield an asymmetric interaction of the membrane with the electrolyte, hence an asymmetric push on the liquid. It Fig. 14 : From nanoscale to angström˙ scale pores. (a) Reproduced was shown in Ref. 50 that such asymmetric geometry – depicted from Ref. 73. (Top) Molecular dynamics representation of water flow- in Fig. 15-a – results in an osmotic diode, with a rectified os- ing through a transmembrane multi-wall boron-nitride nanotube and (bot- motic pressure versus the concentration gradient (non linear de- tom) Transmission electron microscope (TEM) picture of its experimental pendence), furthermore tunable by the applied electric field. counterpart. (b) (top) Molecular dynamics representation of a nanopore in a mono-layer MoS2 membrane (in blue and yellow) and the salt (green The description builds on the previous mechanical views of os- and red) in solution and (bottom) TEM picture of its experimental coun- mosis, in Sec. 2.4. The Stokes equation for fluid motion writes terpart, a 5 nm pore. Reproduced from Ref. 74 with permission from 2 Springer Nature, copyright 2016. (c) (top) Molecular dynamics represen- 0 = ∇p + η∇ u + ρe( ∇Ve), (65) tation of water and ions (orange and blue) flowing through a graphene slit − − with 7 angström˙ spacing. (Courtesy from Timothée Mouterde) (bottom) with ρe = e(c+ c ) the charge density, c is the concentration AFM image of bilayer graphene spacers on top of the bottom graphite − − ± of positive and negative ions (assumed here as monovalent for layer. Inset: Height profiles yield an estimate of 7Åfor the thickness of simplicity) and V is the electric potential. Following the same spacers made from 2 layers of graphene or 1 layer of MoS2 (the blue e line shows the scan position for the corresponding trace in the inset). steps as in Sec. 2.4 to integrate the fluid equations of motion in Reproduced from Ref. 137 with permission from AAAS, copyright 2017. the channel, the general relation between flow and pressure takes the expression Q = L ∆[p Πapp] (66) hyd × − − confinement thicknesses down to 3 Å 142,143 – see Fig. 14-c. ∼ where L is the channel permeance introduced above. The ap- Such nanofluidic technologies offer new possibilities in the con- hyd parent osmotic pressure between the two sides of the channel is text of osmotic transport. They allow nearly molecular scale de- accordingly defined as signs, leading to various nanofluidic-specific effects which may be key assets for new separation techniques and water filtra- 1 ZZ ∆Πapp = dA dxρe ( ∇Ve) (67) tion: from specific ion exclusion effects 87,138,140,143,144 with a A × − number of anomalous ionic effects to be investigated 145, to ex- with A the cross section of the pore, L its length. The ion concen- tremely fast permeation of water, in particular through carbon tration profiles obey the Poisson-Nernst-Planck equations, cou- 139,140,146–148 nanotubes . Also new types of nanoscale mem- pling the diffusive dynamics to the applied electric forces. In spite branes have also emerged recently, offering new designs as com- of the expected non-linear dependence of the osmotic pressure in pared to traditional membranes: for example, with dedicated terms of driving forces, the symmetry in the force balance and 149 patterns of hydrophilic and hydrophobic regions ; or tailor- ionic transport equations, which was highlighted in Sec. 2.4 and 86,150 designed DNA origami channels , and – last but not least – Eq. (28) for the simplest geometry, still holds. There is accord- the multilayer membranes of graphene (so-called graphene oxide ingly a linear relation between the apparent osmotic pressure in membranes) 75,151,152. Eq. (67) and the total surface ion flux js: This constitutes a new and exciting playground, in which osmotic phenomena may (and should) flourish in various forms. L ∆Πapp = kBT∆cs + js . (68) We discuss in the next paragraphs two such examples: the devel- × D opment of osmotic diodes, and an active counterpart of osmosis, It is therefore expected that the rectifying behavior in the ion flux, which both lead to tunable osmotic driving beyond van ’t Hoff. akin to the current diode, thus translates into a rectifying osmotic pressure. 4.2 Osmotic diodes and osmotic pressure rectification Solving the full equations in the geometry presented in Fig. 15- One of the successes of nanofluidics was to demonstrate the pos- a yields a fluid flux: 50 sibility to design diodes for ionic transport, in full analogy with 62,85,153 e∆Ve their electronic counterpart . This takes the form of a non- Q = Lhyd[σkBT∆cs ∆p] + QS[∆cs] exp 1 (69) − kBT − linear and rectified response for the ionic current versus the applied voltage. Typically an ionic diode behavior manifests itself in where the reflection coefficient σ is now a non-linear function of

Journal Name, [year], [vol.],1–43 | 17 (a) 4.3 Towards active osmosis ∆Ve We discuss now a second class of examples of osmotic phenom- - - - + - ena that goes beyond the van ’t Hoff paradigm. As we exhaus- Σ αΣ − + + tively discussed, the idea of osmosis is closely related to semi- + + - + + + + + + + + - - permeability and sieving – with the membrane playing the role of - - - + + + + + + + - - + + + + - a simple colander. However one may consider how the osmotic - - + + + + + + + + + - - - + + + - pressure builds up in membranes with time-dependent pores: a - - + - + + pore which opens and closes over time – see Fig. 16 – will exhibit + Σ αΣ cL − - + cR - - a time-dependent size exclusion and sieving is thus expected to generate an intermittent osmotic push on the ﬂuid. The resulting (b)e (c) osmotic pressure is expected to be some time-average of the push, e e which remains to be properly deﬁned. But injecting energy at the e pore scale – here via the time-dependent opening of the pore – e may also lead to far-from-equilibrium behaviors, allowing possibly to bypass the entropy bottleneck. Osmosis across dynamically stimulated pores is therefore a subtle problem, which requires proper microscopic foundations. Beyond the fundamental question, adding the sieving frequency as a new tuning parameter may improve separation and selectivity properties of the membranes e 158,159.

Fig. 15 : Osmotic rectification in an osmotic diode. (a) A nanochan- (a) (b) small particle large particle nel presents an asymmetric surface charge with Σ > 0 and αΣ on the − with fast diffusion with slow diffusion other side with α = 1. The ions are therefore submitted to an asymmetric 6 force (between one side and the other, in colored arrows) that drives the osmotic flow. Inspired from Ref. 50. Apparent osmotic pressure ∆Πapp FG-Nups versus (b) salinity gradient ∆n = nR nL (where ni = ci/c0 is a normalized − concentration) and versus (c) applied voltage drop ∆Ve = VR VL (normal- − ized by kBT/e) as obtained from an analytical solution for the flows of all species in the nanochannel. (b) and (c) are reproduced from Ref. 50 with nuclear pore complex permission from the APS, copyright 2013. More information can be found low energy barrier high energy barrier in Ref. 50. Fig. 16 : Osmosis through out-of-equilibrium pores. (a) Illustration of a pore- with a time-dependent shape, where the inner pore size may be either smaller either larger than the typical solute size. (b) Repro- duced- and adapted from Ref. 160 with permission from Springer Nature, - the concentrations in both reservoirs and QS plays the role of a copyright 2016. Representation of the spatiotemporal motion of grafted - 50 phenylalanine-glycine nucleoporins (FG-Nups) inside the selectivity filter “limiting fluid flux” . The apparent osmotic pressure ∆Πapp = - of the- nuclear pore complex. Small (resp. large) particles that diffuse fast Q/Lhyd is plotted in Fig. 15. The rectification and diode behav- (resp. slow) see effectively the FG-Nups as static leaving small openings ior versus concentration is weak for zero voltage but strongly en- (resp. moving and everywhere) and therefore encounter only a "small" hanced for higher applied voltage bias. energy barrier for translocation (resp. large).

Examples of permeability rectification are actually numerous in the biological world. They are harnessed e.g. in plant cells 135 or The question of dynamic osmosis actually arises naturally in in animal cells 155–157. Surprisingly in all the studies that we are biological pores, e.g. ion channels, since their shape is affected aware of, entering flows are notably larger than outer flows, and by thermal fluctuations or demonstrates out-of-equilibrium mo- up to 10 times higher in some mammalian fibroplasts 157. It is tion 161. The question of out-of-equilibrium osmosis was dis- fascinating to see how most cells are therefore adapted to fill in cussed in the literature in the 1980s 162 and a few molecular dy- faster than they would swell under the same conditions, probably namics studies were pursued 161,163,164, usually with a focus on with a link to survival strategies. We highlight that rectified os- the specificities of the biological channels under scrutinity. In fact, motic flows could be used in a variety of fields. In fact, Fig. 15-b temporal dynamics of biochannels are strongly believed to be con- and the results reported in Ref. 50, demonstrate that water may nected to the selectivity properties of the channel. For example, be flown against the natural osmotic gradient, with water flowing the fluctuations of the refined structure of the selectivity filter of to the high salinity reservoir, depending on the voltage applied. the KcsA channel is believed to be a key factor for extremely re- Furthermore, under an oscillating electric field, with proper confined passage of the potassium ion 161. Further, such temporal ditions, this induced water flow against the natural osmotic gra- dynamics of the structure may provide an efficient alternative to dient will be maintained. This opens new perspectives e.g. for simple steric sieving for selectivity. This was noticed in the nu- advanced water purification strategies and active filtration with clear pore complex 160, where particles see an effective transloca- oscillating fields, as we discuss later. tion barrier which is dependent on their diffusion properties (see

18 |Journal Name, [year], [vol.], 1–43 Fig. 16-b). (a)(x, t ) (b) (c) effective pump U C2 To our knowledge, there is no general framework discussing free diffusion the concept of “dynamic osmosis”. Several simple models were C1 considered recently by the authors in Refs. 158,159. Here we illustrate a few basic concepts underlying this active osmosis pro- δ0L cess and the opportunities that it offers. (d) k T (C C ) (e) k T (C C ) × B 2 − 1 × B 2 − 1 In line with our previous discussion of osmotic phenomena, it is 2 2 pump particularly fruitful to address the question under the perspective 1 1 app app of the mechanical insights, where the pore with fluctuating shape is modeled as a time-dependent energy barrier, say U (x,t), using ∆Π 0 ∆Π 0 sink similar notations as previously. The average osmotic force acting -1 -1 on the fluid is again obtained in terms of the force acting on the 10 0 10 2 10 0 10 2 fluid integrated over the channel size and averaged over time. It ω/ω0 ω/ω0 writes within this framework Z Fig. 17 : Active membrane as a pump (or sink). (a) An asymmetric ∆Πapp = dxcs(x,t) ( ∇xU (x,t)) t . (70) potential barrier (representing the membrane acting on the solute) sepa- h × − i rates two solute reservoirs with different concentrations. (b) As the barrier is temporarily lowered, the solute may diffuse inwards from both sides. where here t denotes a time average. As for the static (pas- h·i (c) If the barrier is risen back, the solute that has diffused beyond the sive) case, the Smoluchowski equation for the solute allows one maximum point of the energy barrier will be carried to the other side. In to rewrite the apparent osmotic pressure in terms of the solute the example shown here, more solute from the lower concentration side flux across the fluctuating barrier: has traversed beyond the maximal point. This solute is then transported to the highly concentrated side. The active membrane therefore acts as L a pump. (d) and (e) Apparent rejection coefficient as obtained from the ∆Πapp = kBT∆cs + js t . (71) 168 h i × Ds on/off energy barrier model described in the text . The normalizing 2 frequency is ω0 = Ds/L , and the asymmetry parameter is here δ0 = 0.1. It is interesting to note that the concept of osmotic force ∆Πapp Insets indicate the solute concentration on both sides; ∆cs = C2 C1; for − connects directly to the question of translocation of solute (d), C1 = 0.4C0 and C2 = C0 where C0 is some arbitrary concentration and molecules across a fluctuating barrier – via the solute surface for (e) C1 = C0 and C2 = 0.1C0. flux js. That specific question was actually the topic of an exhaustive literature in the context of ratchets, molecular motors, with C and C the solute concentrations in both reservoirs and or stochastic resonance 165,166. Numerous counter-intuitive con- 1 2 δ the potential asymmetry, see Fig.17. Beyond the specific ex- sequences were highlighted, both theoretically and experimen- 0 pression (restricted to this specific regime and model), this re- tally, like directed motion, “uphill” transport against gradients, sult highlights the possibility of uphill solute transport (pumping enhanced translocation, etc. Accordingly the previous symmetry against concentration gradients) or enhanced solute flux, depend- relation Eq. (71) shows that the existence of a finite flux js t , h i ing on the direction of the concentration gradient versus the pore with possibly unconventional dependencies on the concentrations asymmetry. This behavior is summarized in Fig. 17-d-e. in the reservoirs, will have consequences on osmotic transport, i.e. Now inserting this result for the flux in Eq. (71), one therefore leading to flow of the suspending fluid itself and not only solute predicts highly counter-intuitive behaviors for the osmotic pres- motion. sure, i.e. the driving force acting on the fluid. In Fig. 17-d-e, we To illustrate this behavior, it is instructive to consider a sim- have introduced and plotted the apparent osmotic pressure ple example, made of an asymmetric membrane as in Fig. 17, kBTL which oscillates in time as an “on/off” process over a time interval ∆Πapp = kBT(C2 C1) + js t (C1,C2,ω) (73) − Ds × h i τ/2 = π/ω. When “on”, the barrier height is considered as much larger than the thermal energy. This process bares similarities based on the full solution of the simplistic previous model. No- with the ratchet process in Ref. 167 and subsequent references, tably this plot highlights the possibility of “resonant osmosis” for a where solute pumping was demonstrated. In elementary terms, characteristic frequency (in the form of an extremum of σapp), but when the barrier is “off”, solute molecules from both sides diffuse this simple model also suggests that – depending on the direction freely (see Fig. 17-b). Now, when the barrier is back “on”, solute of the potential asymmetry against the concentration gradient – that has crossed the maximum point of the barrier will slide down the rejection may be larger than unity (pumping regime) or even to the opposite side (see Fig. 17-c). This process leads to a finite decrease towards negative values 168. flux of solute averaged over a period, js t , which can be exactly This points to a wealth of intriguing behaviors for osmotic phe- h i calculated in the simple model considered, see Ref. 168. For ex- nomena, which were barely explored up to now. As emphasized ample, in the quasi-static (low frequency) regime, the averaged above, the recent development of nanofluidics suggests many flux reduces in this simple system to routes to develop such active pores in artificial channels, e.g using voltage-gated nanochannels 169–172 or UV light 173 or stimulated js t Lω [C2 δ0 C1 (1 δ0)], (72) surface chemical reactivity 174,175. Other externally controlled ex- h i ω∼ 0 × × − × − →

Journal Name, [year], [vol.],1–43 | 19 176,177 isting devices include thermally responsive nanochannels . (a) c The challenge awaiting is to achieve such active control in yet ∇ ∞ v c DO ∇ ∞ smaller devices to signiﬁcantly impact water or ion transport.

The foundations of active osmosis remain therefore to be properly investigated. The present discussion is merely an appetizer vDO vDP to illustrate the abundance of “exotic” behaviors which could be unveiled in this context. (b) (c) time

5 From diffusio-phoresis of particles to active matter time

The previous sections showed how gradients of solutes induce fluid motion in the presence of an interface via the diffusio- osmotic phenomenon. Symmetrically when a (solid) particle is suspended in a quiescent fluid, gradients of solute will induce Fig. 18 : From osmosis to phoresis. (a) Under a concentration gradi- motion of the particle. This motion, called “diffusio-phoresis”, ent, a particle is put into motion via diffusio-osmosis occurring at its sur- relies on the very same osmotic driving forces, occurring within face. (b) Time-stamped stream lines of decane droplet migration towards the interfacial layer at the particle boundary. The idea to trans- a hydrogel beacon initially loaded with sodium dodecyl sulfate (SDS), port large particles harnessing osmotic forces appeared first in the acting as a long-range solute source. Adapted from Ref. 182 with per- 55,56,99 mission from the United States National Academy of Science, copyright Russian literature with the works of Derjaguin and Dukhin 2016. The scale bar is 100 µm. (c) Diffusio-phoretic transport of fluores- 79,178,179 and was more thoroughly investigated in the 1990s . We cent λ-DNA under a LiCl gradient (∆cs = 100mM over a range of 800µm, refer to the review by Anderson in Ref. 54 for a dedicated dis- highest concentration being up), scale bar is 100 µm. Images at 100, cussion of the underlying transport mechanisms and some of its 150, 200 and 300 s. Adapted from Ref. 77. subtle features.

Diffusio-phoresis and its consequences have gained renewed Eq. (41): interest for the last decade, highlighting an increasing number of 1 Z ∞ U (z) situations where this phenomenon is shown to play a role, as well µDP = µ = z exp − 1 dz. (76) DO η k T as dedicated applications in various domains. Basically diffusio- − − 0 B − phoresis occurs whenever there is a gradient of solute or of a mix- Interestingly, provided the value for the diffusio-osmotic mobility 180,181 ture of solutes and such situations are ubiquitous . Here we is constant over the particle’s surface, it was shown by Morrison summarize the main elements of the phenomenon and focus on that this result holds for any particle shape (the argument is valid a number of elementary implications. More explicit applications for any interfacially driven transport 54,183). will be discussed in the next section, Sec. 6. 5.1.1 Phoresis in the thin layer limit Summarizing briefly the derivation, Eq. (76) is obtained by sep- 5.1 E pur si muove: from diffusio-osmosis to diffusio-phoretic arating the diffusio-osmotic driving, which occurs at the particle motion surface within the diffuse layer of thickness λ, and the far-field flow occurring beyond the diffuse layer. Following the predic- The diffusio-phoretic velocity of a particle under a (dilute) solute tion in Sec. 3.1, the near-field diffusio-osmotic flow results in a gradient writes as 54 diffusio-osmotic “slip” velocity of the fluid relative to the parti- v cle, with amplitude µDO ( kBT∇cs)t , where the index t refers vDP = µDP ( kBT∇c∞) (74) × − × − to the tangential component along the particle surface. The con- Physically the phenomenon at stake is sketched in Fig. 18: the centration gradient in the vicinity of the surface (∇cs)t is related 2 solute gradient at the solute surface induces a diffusio-osmotic to the far concentration field cs. It obeys Ficks’ equation ∇ cs = 0 slip velocity of the fluid (relative to the solid particle) beyond the together with the boundary condition at infinity fixing the concen- interfacial diffuse layer; the particle is put in motion in order to tration gradient, cs(r ∞) z∇c∞, with z the coordinate along the → ' precisely negate the corresponding velocity. For spherical parti- direction of the gradient. This gives (in spherical coordinates): cles, the value of the mobility µDP defined in Eq. (74) identifies 2! with minus the corresponding diffusio-osmotic mobility, as given r 1 R cs(r,θ) = R∇c∞ + cosθ (77) previously in Sec. 3.1: R 2 r ×

µDP = µDO. (75) outside of the diffuse layer; with r,θ the spherical coordinates. − Back to the ﬂow, the velocity ﬁeld outside the diffuse layer obeys For example, for a solute interacting via a potential U with the the Stokes equation particle, the diffusio-phoretic mobility is minus the mobility in η∇v ∇p = 0, (78) −

20 |Journal Name, [year], [vol.], 1–43 with the boundary conditions on the particle given by the tangen- The difference between the two scalings originates in the fact tial slip velocity and at infinity given by the uniform flow field that for interfacially driven motion, the velocity gradients occur mostly over the thickness λ of the diffuse layer, and not on the + 3sinθ v(r = R ) = vslip t, & v(r ∞) = vDP (79) particle size R, as e.g. for the Stokes flow. More fundamentally, − 2 → − this raises the question of how osmotic pressure is expressed: the + in the particle frame of reference; vslip = µDO ( kBT∇c∞), R existence of a difference of solute concentration between the two × − ≈ R + λ denotes the position on the particle surface but located be- sides of the colloidal particle does not obviously imply the exis- yond the diffuse layer (here considered as infinitesimal); t is the tence of a corresponding osmotic pressure and this belief led to tangential vector on the particle’s surface. To lowest order, the much confusion. The argument above based on the global force solution to the previous equations results in a flow dominated by balance is globally flawed and needs to be properly clarified. In a Stokeslet his exhaustive work in Ref. 191 following the debate of Refs. 186– F cosθ 1 190, Brady tackled the question based on a “micromechanical” vr = vDP cosθ + + O − 4πη r r3 analysis of the solute and solvent transport in the presence of the (80) colloidal particles. F sinθ 1 vθ = vDP sinθ + O However, in order to properly solve the riddle and reconcile the − 8πη r r3 various approaches, one needs to go into the details of the force where F = 6πηR(vDP +vslip). As can be easily verified, F identifies distribution and write properly the force balance on the particle with (minus) the force acting on the particle. At steady-state, the undergoing diffusio-phoretic transport. It is accordingly interest- particle is moving with a constant velocity vDP and no force acts ing to relax the hypothesis of a thin diffuse layer, and consider on it. Accordingly, the diffusio-phoretic velocity vDP is fixed by more explicitly the transport inside the diffuse layer, as was ex- imposing F = 0 and this results in Eq. (74). This shows implicitly plored by various authors, using e.g. controlled asymptoptic ex- that the far velocity profile scales like 1/r3 and can be rewritten pansions 193–195. as in Ref. 54 General results for the hydrodynamic flow and mobility can be 3 1 R h rrrr i obtained without assuming a thin diffuse layer. We consider that vl(r) = 3 I vDP. (81) 2 r r2 − · the interaction between the solute and the particle occurs via a radially symmetric potential U (r), so that the Stokes equations where we came back to the lab frame of reference, vl(r) = v(r) − now writes vDP. Accordingly, the hydrodynamic interaction between particles 2 η∇ v ∇p + cs(r)( ∇U ) = 0. (82) undergoing diffusio-phoretic transport is weak, in contrast to e.g. − − gravity driven transport where the fluid velocity scales like 1/r The boundary conditions are now replaced by the no-slip bound- far from the particle. This has important consequences for the ary condition on the particle’s surface, as well as the prescribed phoretic transport in confinement 54,184,185. velocity at infinity (in the frame of reference of the particle):

5.1.2 Osmotic force balance on particles v(r = R) = 0 and v(r ∞) = vDP (83) → − Let us come back to the force balance underlying diffusio- The concentration profile obeys a Smoluchowski equation in the phoresis. We emphasized above that diffusio-phoresis, like any presence of the external potential U (r), in the form interfacially driven transport, is a force-free motion: the particle moves without any force acting on it, i.e. the global resulting force 0 = ∇ [ Ds∇cs + λscs( ∇U )] (84) − · − − acting on the particle vanishes 54. This has counter-intuitive con- with the boundary condition at infinity accounting for a constant sequences and led to various mis-interpretations and debates con- 186–191 solute gradient cs(r ∞) r cosθ∇c∞ (convective transport is ne- cerning osmotically-driven transport of particles , in partic- → ' glected here). Given the symmetry of the problem, the solution ular in the context of phoretic self-propulsion (see Sec. 5.3 and takes the general form c r c r cos , with c scaling with Ref. 192). We thus take the proper space here to discuss the os- s( ,θ) = 0( ) θ 0 the gradient at infinity as c R c . Altogether this is a self- motic force balance. 0 ∝ ∇ ∞ consistent equation for the solute concentration field. It should A naive interpretation of diffusio-phoresis is that the particle therefore be considered as a source term for the fluid transport velocity v under a solute gradient results from the balance of DP Eq. (82). For large distances to the particle (r λ), it reduces to Stokes’ viscous force F = 6πηRv and the osmotic force result- v DP the previous result in Eq. (77). ing from the gradient of the osmotic pressure integrated over 2 the particle surface, hypothetically scaling as Fosm R R∇Π. Interestingly, the solution of Eq. (82) for the velocity profile can ∼ × Balancing the two forces one finds a phoretic velocity behaving be calculated exactly for any radially symmetric potential U (r), by 2 kBT as vDP R ∇cs. Looking at the expression for the diffusio- extending textbook techniques for the Stokes problem in Ref. 96; ∼ η phoretic mobility in the thin layer limit, Eqs. (74) and (76), the see also 196 for a related calculation in the context of electro- latter argument does not match the previous estimate by a factor phoresis. It can be demonstrated that the solution for v(r) still of order (R/λ)2, where λ is the range of the potential of interac- takes the same form as in Eq. (80), but the force along the axis of tion between the solute and the particle. the gradient appearing in the Stokeslet term (v F/r) term now ∼

Journal Name, [year], [vol.],1–43 | 21 takes the expression dients do generate an inhomogeneous local tension on the sur- Z ∞ face of the particle, as plotted in Fig. 19-a. Accordingly, if one 2 F = 6πRηvDP πR c0(r)( ∂rU )(r) ϕ(r)dr (85) considers that the particle is elastically deformable, such tensions − R − × would generate a deformation of the particle following the shape 2 with ϕ(r) = 2 3 r 2 r R a dimensionless function, the sketched in Fig. 19-b. 3 R − R − r 2 factor 3 originating from the angular average. The diffusio- (a) c (b) phoretic velocity results from the force-free condition, F = 0, and ∇ ∞ therefore it writes Force ﬁeld deformation 2 ∞ πR Z Absorbing vDP = c0(r)( ∂rU )(r) ϕ(r)dr (86) sphere 6πηR R − × vDP

Remembering that c0(r) ∝ R∇c∞, this equation generalizes the previous result obtained in the thin layer limit. At first sight, Eqs. (85) and Eq. (86) appear as a force balance between the Stokes friction 6πRηvDP and the osmotic force, here Fig. 19 : Local force acting on a diffusio-phoretic sphere. (a) Local force field acting on a sphere experiencing diffusio-phoresis with absorp- written in terms of the local force c0(r)( ∂rU )(r) integrated over − tion at its surface in a solute gradient. The local force is plotted with an the particle surface (and potential range). The latter represents arbitrary factor amplitude factor (the same for each vector) and projected the push of the solute molecules on the particle. Actually, Eq. (86) in the 2D plane; (b) Potential resulting deformation of the sphere, axisym- is very similar to Eq. (2.7) in Ref. 191, with the r-dependent term metric view, when the deformation is assumed to be proportional to the πR2 ϕ(r) replaced in Ref. 191 by the prefactor L(R). However local force. × the integrated “osmotic push” is weighted here by the local factor ϕ(r) (in contrast to Ref. 191) and this detail actually changes the The situation is very different for electro-phoretic transport. As whole scaling for the mobility. was first discussed in Ref. 197, for electro-phoresis there is a lo- Indeed in the thin diffuse layer limit, with r R λ R, then cal force balance between the direct electric force acting on the − ∼ one may expand ϕ(r) 2(r R)2/R2, while the concentration particle and the hydrodynamic shear acting on its surface: ac- ' − − profile c0(r) can be approximated as cordingly the local force simply vanishes identically. In physical terms, this is due to the fact that the electric force acting on the " 2# r 1 R colloid particle exactly balances the electrical force acting on the c0(r) R∇c∞ + exp[ U (r)/kBT]. (87) ' × R 2 r − electric double layer because of local electroneutrality (the charge in the electric double layer being exactly opposite to the surface One may then verify that the above Eq. (86) indeed reduces to the charge). This can be actually verified explicitly by extending the results in Eqs. (74) and (76) predicted by the thin layer approach. previous calculations to electro-phoresis. This can be performed 2 2 In other words, the weight ϕ(r) λ /R is a signature of the fact 196 ∼ for weak electrostatic potential along the lines in Ref. , and it that the velocity gradients occur on the potential width λ and not predicts indeed a vanishing local force. R on the particle size . An osmotic pressure is indeed expressed at Accordingly, particles undergoing electro-phoresis are not ex- the particle’s surface and yields diffusio-phoretic transport, but in pected to deform, in contrast to diffusio-phoresis which leads a very subtle way which does not reduce to considering only the to local deformations. Such results remain to be experimentally direct solute force. This corrects the naive argument suggested at studied in order to observe the modification of a particle confor- the beginning of the section. mation undergoing diffusio-phoretic drift. We note however that The exact calculation above also allows one to gain key insight in the context of thermo-phoresis, DNA was reported to stretch into the local force acting on the particle. The latter is the sum under a temperature gradient 198. Such effects could have inter- of the hydrodynamic shear force, normal pressure and direct in- esting applications in the context of separation of particles, since teraction with the solute. Using the exact results for the velocity their shape will differ depending on their size. profile in the thin layer regime, λ R, one predicts As a last comment, the previous discussion neglected surface 2 transport at the surface of the particle: this involves convection of fr = 3LsR kBT∇c∞ cosθ solute in the interfacial region, but also fluid slippage at the parti- (88) 3 2 cle surface, which will both affect the steady-state concentration f = LsR k T∇c sinθ θ 2 B ∞ field of the solute around the particle. This leads to corrections to the mobility as a function of a (properly defined) Péclet number, R ∞ βU (x) where Ls = e 1 dx has the dimension of a length and R − − as introduced in Refs. 68,69,193,199. quantifies the excess adsorption on the interface. Eq. (88) can be recovered easily with a simplistic argument: one expects this os- 5.1.3 The diffusio-phoretic mobility motic force to scale as Vint∇Π = Vint∇(kBTc∞) where Vint is the Let us now focus on the mobility. As for diffusio-osmosis, the 2 interaction volume. Writing Ls the typical interaction lengthscale diffusio-phoretic mobility scales as µDP λ /η where λ is the 2 ≈ ± we have V 4πR Ls, leading accordingly to Eq. (88). While the thickness of the interfacial layer. For electrolytes, the latter iden- int ≈ integrated total force does vanish as expected, the osmotic gra- tifies with the Debye layer thickness and one expects accordingly

22 |Journal Name, [year], [vol.], 1–43 that µDP 1/cs so that one usually writes the diffusio-phoretic easily achieved – in relative contrast to electric fields as driving ∼ velocity under salt gradients as: forces – so that this phenomenon can induce particle motion in quite subtle ways leading to a wealth of counter-intuitive behav- vDP = DDP∇ logcs (89) iors. Such solute patterns may occur naturally, for example due to evaporation leading then to drying film stratification 202, in The diffusio-phoretic mobility DDP has now the dimension of a membrane fouling 207, or at dead-end pores, allowing for boosted diffusion coefficient. According to the previous estimates, one extraction of particles in porous media 209,210, as well as in hy- expects for electrolytes that DDP kBT/(8πη`B) with `B the ≈ drothermal pores with steep pH gradients 181. Alternatively static Bjerrum length, so that the value for DDP is expected to be in or dynamic patterns of solutes can be designed thanks to ded- the range – though slightly smaller – of diffusion coefficients icated microfluidic devices 78,203. An illuminating example was of molecules (thus far larger than any colloid diffusion coeffi- reported recently in Ref. 182, showing that “chemical” beacons cient): experimentally typical values for DDP are in the range 10 2 78,200 emitting solutes may allow to engineer ultra-long range nonequi- DDP 0.1 1 10 m /s . Note that, as for diffusio-osmosis, ∼ − − librium interactions between particles, up to millimeters – see diffusio-phoresis under electrolyte gradients with unequal diffu- Fig. 18-b. sion coefficients for the anions and cations (D+ = D ) has an 6 − electro-phoretic contribution similar to Eq. (47) which can be- Instead of being exhaustive, we discuss here several “elemen- come quantitatively predominant; see also Ref. 54. The expres- tary mechanisms”, which serve the purpose of highlighting the sion in Eq. (89) highlights a “log-sensing” behavior, similar to that versatile manipulation of particle assemblies via diffusio-phoretic observed in bacteria, e.g. in E. coli 201. It is at the basis of various motion. unconventional transport phenomena which we discuss below. (a) Colloids (b) Colloids buffer buffer b + salt b + salt Aside the case of electrolytes, other classes of relevant inter- U actions involve steric exclusions – e.g. for neutral polymers – for U U 2 which the mobility is expected to scale as µDP = Rp/η with Rp the excluded particle diameter of the solute 54,115,202. On the experimental side, diffusio-phoresis has been investigated in numerous studies. First measurements were performed 56 z by the Russian school , and later by Prieve, Anderson and collab- (c) (d) orators in the 90’s 178,179. However the development of microflu- M)

idics over the last two decades has allowed to develop dedicated μ systems in which concentration gradients are perfectly controlled

and tunable. It was then possible to measure diffusio-phoretic x/L motion and obtain further insights into the phenomenon and its consequences 21,77–79,200,203–207. While the above discussion assumed implicitly a dilute solute, Dyeconcentration ( the phenomenon is expected to occur under gradients of mix- Distance across channel ( μm) Time (min) tures, and is denoted as solvo-phoresis 79. This was for example investigated in a recent study by Paustian et al. 21, who showed Fig. 20 : Harnessing diffusio-phoresis to transport particles. (a-b) Adapted from Ref. 200. Fluorescent colloids are injected in the central that polystyrene colloids undergo motion in gradients of water- branch of a microfluidic channel with three branches, with the same in- ethanol mixtures. The velocity of the particles was shown experi- let velocity. They are imaged at different positions along the channel mentally to obey (side). In case (a) all channels have the same buffer composition (control), whereas in (b) salt (typically 10 mM of NaCl, LiCl or KCl) is added vDP = DDP∇ logX (90) to the side channels. Although little dispersion is seen in case (a) due to where X is the ethanol molar fraction, thus pointing to non- the low diffusivity of colloids, a strong migration towards the high salinity ideality effects. It would be interesting to disentangle the contri- is observed in (b). The horizontal scale bar is 50 µm. (c) Dispersion of a dye (Rh6G, 50 µM) at a cross-section of a microfluidic system with three bution of the dependence of the interfacial thickness with the mo- branches similar to (a). The dye enters the central channel and the side lar fraction. As a final remark, a slightly different phenomenon is channels are filled with polymer (5 wt % Ficoll 400K) for the spreading ex- the so-called osmo-phoresis, which is obtained for permeable par- periment (control, without polymer). Adapted from Ref. 206 with permis- ticles in which their interface plays the role of a semi-permeable sion from Springer Nature, copyright 2017. (d) Spatio-temporal evolution membrane and reported in Ref. 208. of particles upon exposure to CO2 gradients (CO2 is flown above and below, in x = L). The particles are polystyrene, diameter 0.5µm, dispersed ± in a liquid buffer, and L = 400µm. Adapted from Ref. 205, image under 5.2 Harnessing diffusio-phoresis: membrane less separa- Creative Commons Attribution 4.0 International License. tion, log-sensing and localization In this section we highlight a number of chosen examples to il- Boosting migration – As a first example, we highlight how lustrate the impact and the applications of diffusio-phoresis in diffusio-phoretic transport leads to strongly enhanced migration diverse physical situations. An interesting feature of diffusio- of particles, with the fast solute “towing” the large, slow, par- phoresis is that complex patterns of solute gradients can be rather ticles. This effect was demonstrated in particular in coflow ge-

Journal Name, [year], [vol.],1–43 | 23 ometries, such as in Fig. 20, which was considered in various pa- Ds are the particle and salt diffusion constants and DDP the parti- pers 200,205,206,211. This geometry is quite generic in microfluidics cle diffusio-phoretic mobility. Let us simplify the geometry to fix in the context of mixing and serves here the purpose of highlight- ideas and consider a one-dimensional channel. We consider an ing consequences of diffusio-phoretic motion. Colloids have a low oscillating salt concentration profile, ∇cs(x,t) = f (t)∆cs/`, with ` diffusion constant and therefore barely mix in such a geometry, a characteristic length scale and f (t) an oscillating function of see Fig. 20-a. Adding tiny amounts of salt, typically millimolars, time with zero average. Averaging over the rapid salt concentra- drastically boosts colloid dispersion – see Fig. 20-b-c – with an tion oscillations, the mean diffusio-phoretic velocity which enters observed effective diffusion coefficient of the colloids which is 10 the Smoluchowski equation simplifies to v¯DP = DDP ∇[logcs] t 2 2 h i ≈ to 100 times larger than the equilibrium diffusion coefficient. As DDP( f t /` ) x, where x is the distance to the center of the − h i × mentioned earlier, this is a consequence of diffusio-phoretic mo- cell and t an average over time. It can be rewritten in terms of h·i tion of the colloids under the salinity gradients present across the an effective potential via various parts of the channel. This can be rationalized on the ba- v¯DP µ0 ∂xU , (94) sis of simple arguments. The growth rate for the width of the ≡ × − eff colloid suspension writes as dw/dt = 2vDP = 2DDP∇logcs, with with µ0 = D0/kBT the colloid mobility and cs the inhomogeneous salt concentration. The latter relaxes via 2 Fick’s diffusion and ∇logcs 1/√Dst (the sign depending on kBT x ≈ ± U (x) = f 2 , (95) the salt gradient direction), so that eff 2 t 2 h i σ` dw 2DDP q ` D0 . (91) with σ` = `. This illustrates that the rectified diffusio- dt ≈ ± √Ds t 2 DDP phoresis of the colloids can be interpreted in terms of a harmonic Note that in the experiments of Fig.20, the effective time is related trapping potential towards the central node (x = 0) of the solute to the position z along the channel as t = z/U (U the average concentration oscillation pattern. This allows one to manipulate flow velocity). Integrating this equation yields immediately the the colloidal population via time-dependent solute gradients. observed diffusive like behavior, Alternative routes for focusing colloidal populations were pro- p posed using diffusio-phoretic transport without the requirement w(t) w0 = 2Defft. (92) − ± of time-dependent fields 80 – see Fig. 21-c. These make use of 2 with a diffusion coefficient D D /Ds. Quantitatively D is combined steady gradients of salt and pH which are shown to eff ' DP eff of the order of a (fraction of) salt coefficient Ds so that D yield localization of the particles. The interpretation of this subtle eff D0 (the colloid diffusion coefficient) and colloids “diffuse” much phenomenon incidentally highlights that the effects of gradients faster in the presence of (even minute) concentration gradients. cannot be simply superimposed for diffusio-phoresis and a new Similar behaviors in equivalent geometries have been reported formulation of coupled diffusio-phoretic transport was required, 205,211 206 under CO2 gradients or polymer gradients . rewriting the driving solute gradients in terms of the correspond- Localization – As a second example, diffusio-phoretic motion ing ion fluxes. Such combination of gradients of pH and salts can be harnessed to manipulate and localize particle assemblies. was suggested to occur in hydrothermal pores, with potential Interestingly in the biological world, bacteria are capable of using consequences on the emergence of an ion-gradient-driven early 181 solute contrasts to localize proteins 212. Localization is then used protometabolism and the origin of life . Finally focusing of as an information for further vital processes, for instance localiza- colloidal particles was demonstrated in dead-end pore geome- 210 tion of the ring of the FtsZ protein at midcell is used for cellular tries , with potential applications to preconcentration, separa- division 213,214. Similar features can be obtained on the basis of tion, and sorting of particles. diffusio-phoresis under salt gradients, harnessing the log-sensing Osmotic shock – As a last example, we discuss a very striking feature discussed above in Eq. (89), and leading to particle accu- and counter-intuitive behavior steming from log-sensing, coined mulation 77,78. Indeed, under a linear salt concentration gradient, as osmotic shock, which was discussed in Ref. 77. It illustrates the diffusio-osmotic velocity is not uniform and will be larger in that diffusio-phoresis keeps a long-lasting “memory” of solute gra- regions of lower salt concentration – see Fig. 21-a. Alternating dients, even when they would be expected to be already homog- the gradient over time leads to rectification of diffusio-phoretic enized. Consider a situation where the colloids are spread in a motion and accumulation of the colloids towards e.g.the center reservoir with lateral size `, with initially a uniform solute con- of the cell, as highlighted in Fig. 21-b. It is interesting on this centration c0. Then at time t = 0, solute is flushed at the bound- elementary example to formalize the rectification process. The aries, cs(x = `/2,t) = 0 for t > 0 (simplifying to 1D, and x is the ± colloid and salt equations of transport obey the coupled Smolu- coordinate from the center of the reservoir). After a short tran- chowski and diffusion equations sient, the solute concentration profile will decay to zero according 2 to cs(x,t) c0 exp[ t/τ] cos(πx/`), with τ = ` /Ds the diffusion ' − ∂t ρ = ∇ ( D0∇ρ + DDP∇[logcs] ρ) − · − × timescale of the solute. The diffusio-phoretic velocity of the col- 2 loids then writes vDP = DDP ∇logc(x,t), so that ∂t cs = Ds∇ cs (93) × π πx π2 with ρ the colloid density and c the salt concentration; D and vDP = DDP tan DDP x, (96) s 0 × ` ` ≈ × `2 ×

24 |Journal Name, [year], [vol.], 1–43 dients play a versatile role, readily exploited in many aspects of the biological machinery, such as energy reservoirs, but also serv- ing more sophisticated functionalities, associated with spatial signaling, localization and pattern formation at the various scales involved in the biological processes. One may cite e.g. enzyme transport 217, protein localization in bacteria 212 or more generally in spatial cell biology the use of concentration patterns for po- sitional information 218–220, to quote a few. Obviously chemotaxis in biological organisms under solute gradients is a highly complex phenomenon steming from the interplay of complex signalling pathways, quite far from the simple diffusio-phoretic transport discussed here. But one may reversely remark that the consequences of diffusio-phoresis as a physical phenomenon cannot be overlooked in biological materials, especially in the presence of ubiquitous gradients. This has been barely explored 221.

5.3 From self-propulsion to self-assembly Beyond the idea of passive diffusio-phoresis, where particles move under externally imposed solute gradients, arose the idea Fig. 21 : Focusing particles with diffusio-phoretic transport. (a) that the solute concentration gradients could be generated on the Trajectories of two (yellow) particles starting at different lateral positions particle themselves, e.g. via chemical reactions occurring at their upon an alternating concentration gradient. As the diffusio-phoretic velocity scales logarithmically with solute concentration, the particles closer surfaces. For an asymmetric chemical reactivity, this self-diffusio- to lower concentrations move faster, resulting in localization of the par- phoresis process thus generates self-propulsion of the particles, ticles at the center. (b) Alternating fluids are flushed on both sides of fueled by the chemical reactions 192,222,223. Together with other a circular microfluidic well containing fluorescent colloids. The concen- phenomena leading to self-propulsion, this triggered the emer- tration of LiCl in the two side channels alternates (left/right) with period gence of the field of active matter, which has exploded over the 480s, between buffer alone and 100 mM. The scale bar is 200 µm. (Left) Initial particle distribution in the well and (Right) stationary colloid distribu- last decade. It is not our purpose to review this field, as this goes tion under the alternating concentration gradient. Reproduced from Ref. beyond the scope of our focus on osmotic forces and we refer to 77. (c) Diffusio-phoresis under combined steady pH and salt gradients. some recent reviews on this topic 223–225. We however highlight Flowing NaOH and HCl solutions separately in two reservoir channels here a few phenomena where osmosis, via diffusio-phoresis and establishes a gradient in pH from 3.3 (left) to 10.7 (right), within which diffusio-phoretic particles proceed monotonically to the right. A NaCl related mechanisms, is explicitly at play. gradient is superimposed on the pH gradient inducing diffusio-phoresis Self-diffusio-phoresis – On the experimental side, the phe- to the left. Streamline images reveal unexpected focusing at a location nomenon of self-diffusio-phoresis was pioneered by Paxton and within the channel. Adapted from Ref. 80 with permission from the APS coworkers 226, who showed self propulsion of Platinum/Gold copyright 2016. nanorods in hydrogen peroxide. Hydrogen peroxide is chemically transformed differentially on both metals, either forming or be- pointing towards the center of the reservoir, x = 0, hence gather- ing depleted on each side of the rod, and this creates a gradient ing the colloid population towards this position. From Eq. (96) of the reacting specie (here hydrogen peroxide) driving diffusio- it is clear that the DP velocity is therefore independent of time phoresis, see Fig. 22. For such bimetallic particles with redox ! This leads to the counter-intuitive result that diffusio-phoretic reactions on both sides of the particle, self-electro-phoresis may motion occurs on far longer timescales than the solute diffusion actually contribute to the driving force, via motion of charges timescale. This behavior was highlighted experimentally in Ref. (electrons and ions) within and outside the particle. Self-diffusio- 77, where diffusio-phoretic motion of colloid particles was ob- phoresis was further demonstrated in colloidal janus particles of served on timescales ten times longer than the naive diffusive various materials, see Refs. 227–236, and Refs. 224,237 for a timescale for the salt. Log-sensing is an efficient approach to lo- more exhaustive literature on this aspect. calize colloids, but its application for trapping of other types of On the theoretical side, the mechanisms by which the creation particles, e.g. polymers, remains to be explored. This could possi- or removal of species on the particle’s surface generate an osmotic bly be harnessed to improve sensors or traps for high throughput pressure gradient and motion are not obvious and there has been chemical reactions 215. Applications to information storage and some initial debate on this question, see Refs. 186–190. This retrieval could also be explored 216. Log-sensing also helps re- echoes directly our discussion about the diffusio-phoretic force move particles or fluids trapped in dead-end pores, as we will balance in Sec. 5.1.2. Actually the question was pioneered by discuss in Sec.6.4. Lammert et al. on the putative self-electro-phoresis of biologi- As a final word on this section, the role of diffusio-phoretic cal cells or vesicles driven by non-uniform ion pumping across transport in biological context has yet to be readily explored and the bounding membrane 238. Echoing this situation, an illumi- quantified. In the toolbox of living systems, concentration gra- nating model for self-propulsion via asymmetric osmotic driving

Journal Name, [year], [vol.],1–43 | 25 228,229 (a) v (b) mates D0 , see Fig. 22-b. As a rule of thumb the effective ∞ 2 diffusion coefficient is typically D V τR, where V is the self- eff ≈ × propelling velocity and τR is the timescale for rotational brownian V 1 2 motion: τR D− R /D0, where Drot is the rotational diffusion ∼ rot ∼ coefficient, R the particle size. As a consequence the particles be- H2O2 have as a “hot” bath, with a high effective temperature defined H2O2 from a “fluctuation-dissipation”-like relation as kBTeff = Deff/µ0, where µ0 = D0/kBT the bare particle mobility, so that

Fig. 22 : Self-propelled particles. (a) Chemical reactions occur 2 V τR differentially at the front and at the rear of a reactive colloidal parti- k T k T (98) B eff B D cle, thereby inducing a chemical concentration gradient. This leads to ≈ × 0 diffusio-osmotic driving at the surface, hereby displacing the particle. 2 (up to numerical prefactors). Altogether this predicts that (b) Experimental mean squared displacements ∆L (∆t) and 2D trajec- 2 Teff/T Pe where the Péclet number is defined in terms of the tories (inset) for bare (blue) and active colloids (red) in 7.5% H2O2 solu- ∼ tion. Bare colloids (bottom) show standard diffusion (∆L2 linear in time), self-phoretic velocity as Pe = VR/D0. This prediction was further while the mean squared displacement of active colloids shows a ballistic confirmed experimentally 229. motion at small timescales and a diffusive motion at longer timescales. 2 Osmotic pressure of active suspensions – The question of the os- The measured diffusion coefficients are D0 = 0.33µm /s for bare and 2 motic pressure created by active particles was raised in a num- Deff = 1.9µm /s for active colloids. Reproduced from Ref. 229 with per- 234,241–244 mission from the APS, copyright 2010. ber of theoretical and experimental works . As we introduced above, the osmotic pressure acting on the fluid can be defined mechanically via the average force exerted by the active force was introduced by Golestanian et al. 111,239. They consid- particles on a semi-permeable wall. In the case where the active ered a particle exhibiting a non-uniform chemical reactivity on particles exhibit a Boltzmann-like equilibrium in the presence of its surface, as defined by the corresponding solute flux on its sur- the wall (say, represented by an external potential), as shown e.g. face α(r) = Dr∇ Creact (corresponding to the generation or confor sedimentation profiles 229, then the osmotic pressure reduces − ⊥ sumption of the solute by the chemical reaction), Dr being the to the van ’t Hoff law, except that the temperature is replaced by diffusion constant of the reactant. This boundary term is coupled the effective temperature of the suspension: to the diffusive dynamics of the solute concentration in the bulk. The resulting concentration gradient induces a diffusio-osmotic ∆Π kBTeff ∆ρ (99) ' × slip velocity vDO at the surface, depending on the position, and where ρ is here the concentration of active particles, and the ef- accordingly particle motion. In the case of a janus sphere, ex- fective temperature T was introduced above. This matches the hibiting a contrasting chemical reactivity on its two moieties, the eff equation of state measured using sedimentation profiles 234. self-diffusio-phoretic velocity V takes the simple expression However Boltzmann-like equilibrium is expected to fail in some 1 limiting situations for active particles. In particular it is com- V = vDO sphere = (α α+)(µ+ + µ ) (97) h i 8Dr − − − monly observed that self-propelled particles do hit “compulsively” 240 where α is the chemical reactivity on the two sides and µ the hard surfaces, similarly to a fly on a window : in such cases, ± ± local surface phoretic mobility. We emphasize though that on the particle remains stuck at the membrane’s surface until it re- the experimental side, other mechanisms also contribute to the orients, and there is a non-boltzmannian accumulation of parti- motion of catalytically self-propelled particles, like self-electro- cles at the membrane. This is nicely exemplified in the run-and- phoresis 192. tumble model under an external field, where strong deviations To some extent, such "active particles" mimick self-propelled from the Boltzmann profile is predicted when the typical drift ve- biological organisms. The possibility to fabricate artificially these locity Vd = µ0Fext (with Fext the maximum external force, say, systems constitutes a playground to study far-from-equilibrium due to the separating membrane), is larger than the particle self- 245 behaviors 224. Because active particles consume energy at a local propulsion speed V . This situation occurs for steep potentials. scale, their collective behavior is a priori not constrained – at When such accumulation occurs, the corresponding osmotic push least to some extent – by thermodynamics and may possibly allow will differ from the simple van ’t Hoff law, see Refs. 243,244. In to break the bottleneck of the second principle; cf. the beautiful particular, the osmotic pressure depends on the properties of the example of the rotating Feynman ratchet with active materials in membrane itself and its interaction with the particles – typically Ref. 240. via the ratio between the typical membrane characteristic steep- 244 Active suspensions – Such active particles have fascinating be- ness and the particle mean-free path –, in strong contrast to haviors, and we focus on a few examples. First, self propulsion the van ’t Hoff “universal” relation. Similar deviations from the leads to ballistic motion on short timescales, but orientational van ’t Hoff law also occurs when the interaction between the ac- random motion leads to diffusive behavior on long timescales, in tive particles and the membrane involves wall-induced rotational 242 a way similar to the so-called “run and tumble” motion of bacte- torques . ria. The effective diffusion coefficient is however far larger than Towards self-organization and self-assembly – Another interest- the bare diffusion coefficient based on the Stokes-Einstein esti- ing feature of particles propelling via self-diffusio-phoresis is that

26 |Journal Name, [year], [vol.], 1–43 they interact via chemical signaling. Propelled particles act as a This highlights that the “osmotic interaction” via the diffusio- beacon – similarly to the situation considered in Ref. 182 – and phoretic motion induced by the solute traces leads to long range leave a “trace” of their passage in the form of a diffusing cloud non-equilibrium interactions. This is expected to lead to strong of chemicals which will be felt by other paticles, see Fig. 23. collective effects, such as the clusterization observed experimen- Accordingly other active particles will be reoriented towards or tally. For attractive systems, µDP > 0, these equations are for- against 246,247 the active particle via diffusio-phoretic motion (on mally analogous to a (non-inertial) gravitational system. Keller- top of their self-driving motion). Indeed the surface creation Segel and subsequent works have shown that the above equations or consumption of solutes generates long-distance distortion of predicts an aggregation mechanism, similar to a Chandrasekhar the solute concentration profile, typically relaxing spatially as a gravitational collapse 249,250. Similar conclusions were predicted 251 monopole δcs 1/r (the scaling deriving from Fick’s equation for thermally active colloids where the threshold for collapse ∼ with a sink). This long-range interaction is for example high- was rederived. lighted in Fig. 23-b, showing the diffusio-phoretic attractive motion induced by a single beacon, from Ref. 231. At shorter range, the interaction may become more complex and requires detailed investigation of the chemical drivings 247. This osmotic-induced chemical interaction is at the origin of many advanced collective properties of active particles, such as clustering 224,230,232,248, or self-assembly 231,235, see Fig. 23-c-d. Out-of-equilibrium self-assembly has raised enormous interest, since activity leads to unexpected structures, with the hope of designing novel and smart materials 224. It is interesting to formalize the basics of the phenomenon at stake by writing the coupled diffusion-reaction equation for the colloid population and solute concentration. For the purpose of illustration, we only consider here a single neutral chemically generated specie which acts as a chemo-attractant to the colloids. Fig. 23 : Out-of-equilibrium self-assembly. (a) Sketch of the dif- These dynamical equations actually identify with the so-called fusing chemical trace left behind active particles, and which modifies Keller-Segel equation, which were written to describe the chemo- the local chemical gradient. Another active particle approaching this tactic aggregation of a slime mold (amoebae) under the perspec- chemical gradient will therefore sense a different driving velocity along tive of a dynamical instability 230,249,250: its edge, changing its trajectory (here in an attractive configuration) (b) Courtesy from Jeremie Palacci, data from Ref. 231. Passive colloid ∂t ρ = ∇.( D ∇ρ + µDP∇cs ρ)) diffusio-phoretic speed as a function of the distance to a hematite cube − − eff × (100) that can be used – with blue light – to catalyse the dissociation of H2O2. 2 (c) Spontaneous self-assembly of active particles. The particles form var- ∂t cs = Ds∇ cs + αρ 0 ' ious cluster sizes and shapes. The scale bar is 10µm. Reproduced from Ref. 230 with permission from the APS, copyright 2012. (d) Sequen- with D the effective diffusion coefficient of the active colloids, eff tial self-assembly of DNA-grafted droplets, the different colors represent Ds the diffusion coefficient of the “chemo-attractant” specie and different functionalizations. The scale bar is 10µm. Reproduced from α the chemical rate of the powering chemical reaction occurring Ref. 252, image under Creative Commons Attribution 4.0 International at the surface of each colloid; we assume here that the solute dy- License. (e) Targeted assembly of phototactic swimmers into nanogears. namics are fast. By analogy to electrostatics, the second equation The scale bar is 1µm. Reproduced from Ref. 235 with permission from Springer Nature, copyright 2018. for the solute allows one to obtain the solute concentration as a R function of the colloid density, as cs(r) α/Ds drr ρ(r )/4π r ' 0 0 | − r . When introduced in the first equation in Eq. (100), this shows However the clusters observed in the experiments, e.g. in Refs. 0| that the diffusio-phoretic attraction acts as a self-consistent effec- 230,231,248, do not correspond to full collapse and are rather dy- tive potential such that namic, with clusters reaching a finite size and continuously rear- ranging over time, see Fig. 23-c. As shown in Refs. 253,254, this Z µDP α ρ(r0) behavior can be reproduced by Keller-Segel-like dynamics pro- Ueff(r, ρ ) = drr0 . (101) { } − µ0 Ds 4π r r | − 0| vided both translational and rotational phoretic conditions are properly taken into account in the kinetics. Using this frame- with µ0 the bare colloid mobility. Equivalently, the dynamics of the colloid population can be formally derived from an effective work, it is furthermore possible to predict the condition in which 253,254 free energy functional of the colloid system which takes the sim- dynamic clusterization occurs , in good agreement with the ple form experiments. Beyond clusterization, the self-diffusio-phoretic motion of par- Z Feff = kBTeff drr [ρ(r)logρ(r) ρ(r)] ticles and their osmotic interactions were shown lately to lead − to the self-assembly of active particles into higher levels of struc- Z ture organization. This is highlighted in Fig. 23-e, from Ref. 235, 1 µDP α ρ(r0)ρ(r00) drr0drr00 (102) − 2 µ0 Ds 4π r r where self-spinning microgears are built on the basis of these non- | 0 − 00|

Journal Name, [year], [vol.],1–43 | 27 equilibrium interactions. Beyond, more "on demand" structures in order to exceed the osmotic pressure. are possible, like structures assembled through DNA-grafted in- In a different approach, forward osmosis (combined with ther- terfaces 252,255,256 (see Fig. 23-d). mal methods for desalination) makes use of draw solutions to counterbalance the salinity induced osmotic pressure 270–272 – see 6 Osmosis, towards applications Fig. 24-b. Generating a high osmotic pressure, typically above the 30 bars of pressure between sea and fresh water, requires draw From food processing in biological organisms 257–259, to reverse solutes which are highly soluble in water, and also with a suf- osmosis for desalination and energy generation from salinity dif- ficiently small size (hence low molecular weight). Indeed, as a 31,73,260,261 3 ferences , osmotic forces are harvested in a consider- rule of thumb, for a solute with elementary volume v0 r with ∼ able number of applications in very different domains. In this r the solute size, the maximum osmotic pressure which can be section we review more specifically a variety of such applications achieved is typically ∆Π kBT . This would suggest that solutes ∼ v0 based on (recently) elucidated transport mechanisms relying on with size above 1 nm are not able to achieve a sufficiently high osmotic forces. osmotic pressure for desalination (note that this argument forgets departures from ideality, which could increase more strongly the 6.1 Water treatment and membrane separation osmotic pressure). In a more subtle way, a solute with a strong 6.1.1 Reverse and forward osmosis and their limitations affinity towards water may also decrease the water chemical potential and modify accordingly the pressure. This echoes the huge Access to clean water and cleaning water from industrial waste is pressure drops measured with hydrogel structures 273. a great challenge 262: in 2015 still 663 million people worldwide A number of other membrane based techniques with similar ge- lacked access to drinkable water 263, and cleaning waste water is ometry are used or being developed, from electrodialysis (based becoming a major challenge in oil and gas industries 264,265; go- on electric potential driving of salts) 274,275 to capacitive deion- ing further, some new regulations may appear to enforce a zero ization 276–278, but also shock electrodialysis based on the idea liquid discharge for industrial waste, thus requiring complete re- of combining salt recovery with a porous charged material 279, cycling of water ressources 266. On a day-to-day basis, humanity concentration polarization 280, and other techniques harnessing consumes the equivalent of 10-100 cubic kilometers of fresh wa- chemical phenomena like adsorption desalination 281 or biode- ter 267 for all purposes (agriculture, industry, domestic). Because salination 282,283 and bio-water treatment 284. fresh water is not directly accessible everywhere, and in order to Membrane-based technologies suffer from a number of limita- cover the growing need for freshwater, desalination of sea water tions. First, they have a high-propensity to fouling by molecules and cleaning of waste water have become essential. which are larger than the critical molecule size allowed to pass 285; also due to the large pressures applied during reverse (a)pP p p p 0 (b) Osmotic activator ≫ osmosis. Second, because they are passive membranes – essentially discriminating particles upon their size – they can not be at the same time very selective and highly permeable. This was formalized for ultrafiltration membranes in Refs. 286,287. In fact increasing the permeability of a membrane (and therefore the energy required to recover a given amount of cleaned water) requires essentially to broaden the size of the pores, as water flow Fig. 24 : Reverse Osmosis and Forward Osmosis. (a) Schematic within pores is limited by friction on the pore walls. However explaining reverse osmosis, occurring via a piston applying a large hy- this leads inevitably to a decrease in the selectivity or separation drostatic pressure p such that the difference to the atmospheric pressure properties of the membrane; and reciprocally. This is called the (pressure of the other compartment) is larger than the osmotic pressure selectivity-permeability trade-off. For nanofiltration membranes p p0 ∆Π. (b) Schematic explaining forward osmosis, occurring via the − ≥ addition of some soluble species (here the purple solute) that increases (used for reverse osmosis and so on) the same trade-off exists the osmotic pressure on the drought side and therefore "attracts" water although the proper establishment of a limiting regime is still em- from the brine side. pirical 261 – see Fig. 25. Finally, one challenging progress route for membrane separation is the ability to perform molecular scale Lately, membrane based technology has established itself for design 288 and therefore to ensure the best selectivity properties water purification. Reverse osmosis is the most broadly used tech- to eliminate e.g. micropollutants, some of which are of great con- nique (representing 62-65% of the installed capacity in 2015 for cern for health 289. Overall, it should be realized that the main desalination 268,269, 24% being thermal based technologies). Re- current challenge in desalination and water purification is not reverse osmosis relies on the very simple principle of applying an ally the permeability of the membrane, but rather achieving a external large hydrostatic pressure to counterbalance the osmotic well-controlled selectivity to retain/reject specific species. pressure difference and induce a flow of water towards the low 6.1.2 What can we expect from new nanomaterials and concentration side – see Fig. 24-a. In particular, one can there- nanofluidic devices ? fore extract water from seawater by concentrating seawater even more, or extract water from waste water (in a simplistic view). It may appear that developing the “ideal membrane”, which is For desalination the pressures involved are typically of 30-50 bars both highly selective and highly permeable, is like squaring the

28 |Journal Name, [year], [vol.], 1–43 (a) 10 4 (b) ionic separation 301 and may well revolutionize the domain of ﬁl- 10 -3 tration. ) 3 10 -1 Now, beyond materials themselves, it should be realized that nanoscales allow for many new “exotic” transport phenomena, 10 2 Selectivity permeability 10 -4 the consequences of which have – up to now – been barely har- trade-off

Separation Factor 10 1 nessed. One may quote for example the membranes made of hy- Water selectivity (Pa Water drophobic nanopores, making use of nanobubbles as a semiper- 149 10 0 10 -5 meable sieve for osmotic phenomena ; or the ionic and os- 0 2 4 6 8 0.01 0.1 Water Permeability (10 -9 m/s.Pa) Water Permeability (10 -9 m/s.Pa) motic diodes, allowing for rectified transport in membranes, or the active osmotic phenomenon, as we discussed above; or in a Fig. 25 : Selectivity Permeability Trade-off. (a) Adapted from Ref. different context, the specific adsorption properties of graphene 286. Selectivity versus permeability values for ultrafiltration membranes oxide membranes allowing for water-ethanol separation in mem- (used for separation of larger molecules than salt ions). Bovine serum branes 302, which are far more efficient than standard distillation. albumine is the model molecule for selectivity. The line indicates the standard selectivity permeability model (with a log normal distribution of Such routes would deserve a proper exploration to go beyond the pores, Poiseuille flow and selectivity given by Zeman and Wales exclu- relatively basic sieving principles underlying membrane science. sion rules 290). (b) Adapted from Ref. 261 with permission from Springer These should offer alternative routes for filtration and separation Nature, copyright 2016. Selectivity versus permeability values for reverse which still need to be invented. osmosis membranes using salt as the model specie for selectivity. The lines correspond to the empirical models inspired from Ref. 291 to relate maximal selectivity and permeability in reverse osmosis membranes. 6.2 Osmosis in biological systems: aquaporins, ion pumps and the kidney Osmotic forces are harvested in the biological world in a con- circle. However, Nature has achieved this tour-de-force, with wa- siderable variety of phenomena and contexts: to store energy, ter porins like aquaporins exhibiting unrivalled performances in induce mechanical motion, control ejection and absorption of 292 terms of selectivity and permeability . This requires to develop compounds, etc. Osmotic pressure was much studied at first new artificial materials with properly decorated nanopores allow- in plants 5, and one may in fact assess that plant life depends ing for such exquisite design, for example self-assembled artifi- on osmotic forces. Indeed plants can not rely on muscle for 292–294 cial water channels , or tailor-made DNA origami chan- force generation, yet they are able to achieve tensile and com- 86,150 nels . This is actually a challenge that nanoscale science may pressive stresses on a much wider range 303. To produce mo- 88,152,292,295 be up to . tion or growth, they rely on an underlying hydraulic machinery The development of new nanomaterials has indeed allowed the driven by osmotic or humidity gradients. For example, phloem † emergence of new avenues for membrane separation. Graphitic flow harnesses osmotic driving to transport sugar over long dis- materials of various forms and geometries, such as carbon nan- tances 258,259,304–306. Osmotic transport is critical to regulate size otubes, graphene and lately graphene oxides membranes, have in e.g. conifer leaf 307. The opening and closing of stomata on raised considerable promises, see Refs. 88,152 for reviews on this 308 leaves ‡ and the circadian motion of various plants and flow- topic. Carbon materials were consistently shown to exhibit ultra- ers 309,310 is regulated by swelling or shrinking driven by water low water friction and high permeability, and this represents a flows. Those flows are generally actuated by active transport of key asset to minimize the viscous loss in separation processes. solutes through specialized pumps 311,312. Even biofilms harvest Furthermore advanced functionalization allows one to decorate osmotic pressure gradients in the extracellular space for surface 72,296,297 the nanotubes improving selectivity . Membranes made motility 313. of nanopores in e.g. 2D graphene sheets have a molecular thick- In animals and human beings, a number of processes involve ness, while keeping high mechanical strength: this accordingly osmotic flows for water or volume regulation and transport: from increases the driving forces for transport (which scale like the the kidney 314 to the liver 315 and the intestine 316, not forgetting inverse thickness) by orders of magnitude, hence all transport co- salivary secretion 317. Cells harvest osmotic forces in a variety 88 efficients and the overall efficiency of the process . of ways, most obviously to control expansion and regulate size, Still, these graphitic systems – carbon nanotubes, graphene slits e.g. in cysts 318, and also to regulate absorption 319 or ejection and more generally 2D materials – remain difficult to fabricate of genetic material 320 via small capsules. A number of processes as large-scale membranes. Upscaling towards industrial applica- also harvest more subtle forces in a fascinating way and we cite tions is a considerable challenge. The advent of graphene oxide a few to engage the curious reader. Osmotic pressure changes membranes and their derivatives may change the story. These are may affect frequency of miniature end-plate potentials in neuro- constituted of graphene flakes, which organize into parallel stacks muscular junctions 321,322, but also drive oscillatory flows for cell of graphene layers, having nanoslits in a staggered alignment and regeneration 323. Electro-osmosis is harvested for uphill transport an interlayer distance which is typically below the nanometer. In spite of the complex labyrinthine flow across multiple graphene layers 298, the membranes demonstrate large permeability 299,300. Phloem is a living tissue that transports soluble compounds in particular sugar in † Last but not least, they are relatively easy to fabricate at large plants scale. Such systems therefore appear as ideal membranes for Stomata are small pores at the leaves surface that control leaf transpiration ‡

Journal Name, [year], [vol.],1–43 | 29 of water by insects in draught areas 324 but also more generally they are hydrophobic channels 338 and they are extremely confor epithelial transport 325,326. stricted 338 – only 3 Å in diameter at the narrowest point that The list of examples of osmotic transport in biological systems allows for this selectivity. An aquaporin-based membrane con- is nearly infinite, and occurs at all possible scales from individ- stitutes therefore a somewhat ideal semipermeable membrane. ual molecules to organs and tissues. It is pointless to attempt a All of its exceptional transport properties are intimately con- thorough review. Rather, we discuss below in more detail three nected to its nanoscale (and even Angström-scale)˙ structure – specific biological phenomena related to osmosis. These examples thus hinting to the striking and appealing properties of fluid raise in particular the question of whether such phenomena may flow at the nanoscale. It is thus natural to look for artificial so- be mimicked artificially to achieve advanced osmotic transport in lutions for semi-permeable membranes harvesting properly de- artificial devices. signed nanoscale structures 293,294. For example, aquaporins present a sophisticated hourglass shape, that is believed to en- 6.2.1 Aquaporins: the ideal semi-permeable membrane hance the water permeability 339, see Fig. 26-b. Such a geometry A decisive turnpoint in the study of nanoscale systems was trig- could be readily mimicked using e.g. pore coatings 340 to enhance gered notably by the discovery of nanoscale channels in biology. permeability of membranes. One of the most famous of these channel families is the aquaporin family (the most common being AQP1 or CHIP-28, see Fig. 26- 6.2.2 Kidney: an ultra-efficient and unconventional osmotic a) 327,328. An aquaporin is a water-specific channel; aquaporins exchanger are present in many organs in living systems, animals, but also As a second example, we discuss the separation process occur- plants 329: they play a central role in the human kidney (see be- ring in the kidneys. As we highlight, the efficiency of the kidney low), are also key role players in red blood cells and many other filtration process takes its root in a very unconventional osmotic organs 330, and regulate water uptake in plants 331. The strik- process, and could be inspirational for future separation technolo- ing specificity of aquaporins is that they are both highly selective gies. to water and highly permeable. The permeability of an aqua- Per day, the human kidney is capable of recycling about 200 L porin was measured notably by P. Agre et al. to be in the range of water and 1.5 kg of salt, separating urea from water and salt 14 3 1 332,333§ of p f = 11.7 10 cm .s− at 37◦ C (with p f here de- at the low cost of 0.5 kJ/L 341 while readsorbing 99% of the × ' fined as Q = p f vw∆p/kBT; Q is the water flux and vw the bulk water input. The core of the kidney separation process lies in the water molecular volume). This corresponds to 3 million water 341 ≈ millions of parallel filtration substructures called nephrons .A molecules translocating per second per bar (p f being related to striking feature is that the nephrons of all mammals present a the particle flux dN/dt according to dN/dt = (p f /kBT)∆p). The precise loop geometry, the so-called Loop of Henle - in the shape value of the permeability of the aquaporin is much larger than of the letter "U" – see Fig. 27-a. This loop plays a key role in 292 that for other channels, see for a comparison, or what would the urinary concentrating mechanism and has been extensively 62,335 predicted by continuum dynamics at these scales . studied from a biological and physiological point of view 341–345. As put forward in Ref. 314, the U loop acts as an osmotic ex- (a) (b) Two vestibules changer, similar in concept to a thermal exchanger – see Fig. 27-a. a 16 = entrance effect A mixture of water, salt and urea (or any other compound to be 14 separated) enters and flows through the tubular U loop. Water 12 10 and ions may be exchanged through the tube walls with a com- 8 mon interconnecting media, called the interstitium. On the de- 6 scending side (D), aquaporins allow for water permeation across GlpF 4 AQP1 the walls. On the ascending side (A), salt is actively pumped, Effective pore diameter (Å) pore Effective 2 using an external source of energy (in the case of the kidneys, 0 0 10 20 30 40 50 60 70 the dissociation energy of Adenosine Tri-Phosphate, ATP). This Single file region Relative position (Å) pumped salt results in an increased salt concentration in the in- = selectivity terstitium, higher than the concentration of salt and urea in the Fig. 26 : The aquaporin AQP1 water channel. (a) Molecular dynamics descending tube (D). The osmotic pressure is therefore inverted simulation of water transversing an aquaporin channel. Snapshot from and drives water from the U tube to the interstitium across the movie in Ref. 336 under Creative Commons license, in complement to aquaporin channels. As a result, urea is highly concentrated in Ref. 337. (b) Effective pore diameter of the AQP1 and GlpF channels. the U loop, while salt and water are redirected from the intersti- Pore diameters were determined with AMBER-based van der Waals radii and analysed using the program HOLE38. Reproduced and adapted tium towards the blood circulatory network. This U-shape geo- from Ref. 338 with permission from Springer Nature, copyright 2001. metrical design is key to the efficient operation of the separation. Note that the third limb following the U-tube plays a crucial role 314 Aquaporins present several intriguing features: surprisingly in enhancing the separation efficiency . One may actually estimate the working efficiency of this osmotic exchanger in a simple way, providing a lower bound on

§ A more recent measurement in Ref. 334 suggests that this value may actually have the separation ratio. It is quantiﬁed in terms of the amount of A,top D,top been underestimated by a factor 5. lost water η = cw vA,top/cw vD,top, where v is the ﬂow veloc-

30 |Journal Name, [year], [vol.], 1–43 (a) (b) 10 1 geometry with 3 limbs that allows for a more efficient reabsorp- semi- permeable tion of water. membrane D+A D I A I CD In fact, energy wise, this system is also shown to be far more W) 0 µ 10 salt pumps ( RO efficient than standard reverse osmosis principles, as can be esti- (active ) η mated within the above model, see Fig. 27-b. In living systems, Power D+A+CD the nephron operates the separation of urea from water near the water 10 -1 thermodynamic limit, 0.2 kJ/L 314. Yet, standard dialytic fil- osmotic ' activator Thermo tration systems, which are based on reverse osmosis and passive -2 -1 equilibration with a dialysate, require more than two orders of waste 10 10 Water loss ratio η magnitude more energy 346. Some attemps to build artificial devices mimicking the nephron Fig. 27 : The osmotic exchanger principle of the kidney. (a) Inspired from Ref. 314. Water, salt and urea molar fractions are represented in were reported in the literature, but they rely on biological tissues various colors along the U tube (descending, D and ascending A) limbs or cell mediated transport, and cannot be easily scaled up and and the interstitium (I). For visibility, the water molar fraction was divided transferred to other separation devices 347–349. None of the ap- by 100. Black arrows represent the direction of flow. A semi-permeable proaches so far rely on the specific geometry of the U-loop to membrane (containing aquaporins) separates the descending limb and the interstitium, while the ascending limb contains salt pumps transport- improve the filtration process. Mimicking the separation pro- ing actively the salt to the interstitium. A third limb, the collecting duct cess occurring in the kidney based on the physical perspective (CD, in lighter collers) also exchanges with the interstitium via a semi- described above can now be foreseen using microfluidic elemen- permeable membrane. The latter is crucial for the overall efficiency of tary building blocks. Such an artificial counterpart could have the separation process. (b) Adapted from Ref. 314. Power required for definite impact in improving the early stage of desalination pro- the functioning of the separation process as a function of the targeted water loss ratio: for the simple loop geometry (A+D), for the full serpen- cesses consisting in water pre-treatment to suppress large pol- tine (A+D+CD), as compared to the equivalent reverse osmosis process luting molecules, and, down-the-road, could lead to substantial under a pressure gradient (RO). (a) and (b) are under Creative Commons progress in dialysis for patients suffering from kidney failure. Attribution 3.0 License. 6.2.3 Proton pumps, chemi-osmosis and advanced ionic machinery ity calculated at the top of the ascending (A) or descending (D) As a last example, we discuss proton pumps and channels, which branch, and cw is the concentration of water. For the system to are compelling illustrations of how Nature harvests osmotic forces work, water has to flow from the descending branch towards to drive mechanical parts. Biological systems have developed a the interstitium and this requires that chemical activities obey fascinating artillery of devices to passively and actively transport aWater aWater. The latter can be expressed simply (in the low D,top ≥ I,top ions, namely ionic channels and ion pumps. Among these, proton concentration regime) in terms of molar fractions and one obtains pumps are canonical examples. We detail a few examples below. D,top I,top Proton pumps to build proton gradients – There exists a great va- cw cw (103) riety of ways to actively transport protons in biology, from com- cD,top + cD,top + cD,top ≥ cI,top + cI,top w s waste w s bined proton-electron transfer in cytochrome oxidase (crucial for 350 where cw, cs and cwaste are respectively the concentrations of wa- respiration ) to proton pumps implying the participation of ATP ter, osmotic activator (salt) and waste. Assuming that all the os- – the latter are called H+-ATPases 351. ATP-ases play a key role motic activator has been reabsorbed in the upper branch yields in bio-energetics and are ubiquitous in many forms of life and I,top vD,top D,top D,top 352 cs = cs . Water flow is conserved and thus cw v = plants . They include three types. The P-type ATP-ases include vI,top D,top I,top A,top in particular the plasma membrane H+-ATPase, that uses the dis- cw vI,top + cw vA,top. A lower bound for the fraction of lost wa- sociation energy of ATP to form gradients of protons. These gradi- ter ηlost can then be simply deduced from Eq. (103) as ents are crucial for plant movement (from phloem loading, to size D,top !n 353 c regulation in the stomatal aperture, to tip growing systems ¶ ). η waste . (104) lost ≥ D,top D,top The V-type ATP-ases also use the dissociation energy of ATP to cs + cwaste form gradients of protons. Interestingly this chemical reaction is with n = 1. For the geometry including a third reabsorbing accompanied by a rotary motion of the protein. It is central to branch, the collecting duct, see Fig. 27-a, a similar reasoning many processes in animals 354, from acid base balance in the kid- yields the same result with n = 2. The square exponent thus leads ney, pH maintenance in mechanosensory hair cells, bone resorp- to much smaller lost water fraction ηlost showing that this third tion, tumour metastasis, sperm motility and maturation etc. The branch is essential in the overall efficiency of the kidney separa- last type, the F-type, can work similarly to the V-type 355,356 and tion. Using physiological values for the concentration, this esti- consumes ATP to form gradients of protons depending on aero- mate provides a prediction for water reabsorption, and thus urea bic conditions 357. However it most commonly works the reverse separation, in the range of η 1%, which is in excellent agree- lost ∼ ment with every-day life experience; see 314. To some extent, note that the osmotic exchanger of the kidney may be compared Tip-growing systems, such as pollen tubes or root hairs, continuously grow in one ¶ to a forward osmosis process. However the key difference is the direction.

Journal Name, [year], [vol.],1–43 | 31 way, e.g. consuming the proton gradient and synthesizing ATP, are still investigated 366. Further physical insight on the detailed and we discuss that below. flows in the proton pump but more broadly on ionic channels is required to establish biomimetic principles to construct similar (a) + (b) flagellum H+ H ionic machines with artificial material. Such physical insight is Flow outer membrane rotation also dependent on better modeling of ion transport at the ulti- intermembrane of protons H+ space mate scales, with strong charge interactions, breakdown of hy- H+ Flow of protons drodynamics, etc. cytoplasm stator H+ H+ 6.3 Blue energy harvesting: osmotic power and capacitive mixing ATP synthesis inner membranene As we have seen, filtration and separation of molecules requires rotor turning H+ energy input to counteract the entropy of mixing. Reversely, en- ADP P tropic energy harvesting may be possible by mixing molecules. The energy harvested from differences in salinity, e.g. by mixing sea Fig. 28 : Harvesting proton gradients: energy vectorization and lo- water and fresh river water, is called blue energy. The maximal comotion (a) Simplistic view of an F-type H+-ATPase, here working as entropic energy collected by mixing volumes of sea and river wa- an ATP synthesis enzyme. A proton gradient is maintained between the 3 intermembrane space and the cytoplasm by the respiratory cycle. Pro- ter is typically 0.8 kWh/m , see Ref. 367. Over the earth, count- tons thus naturally flow inwards through the proton channel of the ATPase ing the natural potential resources where rivers flow in the ocean (in yellow). This triggers a mechanical rotation of the central element of such as the Amazonian river, a total of around 1TW of power the ATPase that in turn catalyzes the synthesis of ATP from Adenosine could be harvested, amounting to 8,500 TWh in a year 260. This diphosphate (ADP) and Phosphate (P). (b) Simplistic view of the bacterial flagellar motor (in blue). The proton gradient transverses here the stator is to be compared with the actual production of other renewable parts of the motor (in yellow), namely the MotA/MotB complexes. These energies: in 2015, hydraulic energy production is 4,000 TWh, ∼ are responsible for turning the basal rotor of the flagellar motor. As the the nuclear energy around 2,600 TWh, and wind and solar 1,100 flagellum is attached to the motor, this induces rotation of the flagellum TWh altogether 368. In the global energy balance, blue energy, and allows for bacterial locomotion. as a renewable and non-intermittent source of energy, has thus a great potential. Here we focus on some energetic and osmotic Proton gradients harvested for energy vectorization and locomotion aspects of blue energy and refer to Ref. 75 for a more detailed – The idea that osmotic gradients could be harvested for advanced review of the current status of blue energy harvesting. functionalities was introduced as early as in the 1960s, by the seminal work of Mitchell in Ref. 358. He introduced the concept (a) (b) of chemi-osmotic coupling, namely that a chemical reaction may be e- e- powered by the directed channeling of a specie. In the case of the + electricity turbine anode cathode F-type ATP-ase, directed motion of protons (note that the full reac- - - tion does imply the production of water on one side of the mem- + - - + - + + 356 + - + - brane) catalyzes the synthesis of ATP through a rotary motion - electricity + - + - - + – see Fig. 28-a. As the F-type catalyzes the formation of ATP (that + + - - - - + is the vector for energy in all living systems), it is central to all + + + - + - forms of life 359. The rotary motion occurring during synthesis can be harvested for artificial locomotion of inorganic devices 360 semi-permeable anion selective cation selective membrane membrane membrane fresh sea fresh sea fresh . Harvesting proton gradients for locomotion is more commonly water water water water water performed not by the F-type ATP-ase but by the bacterial flagellar motor 361 – see Fig. 28-b. The bacterial flagellar motor is an 361 Fig. 29 : Collecting blue energy. (a) Pressure retarded osmosis (PRO). impressive 45 nm ionic machinery at the root of bacterial lo- The mixing of sea water and fresh water across a semi-permeable mem- 362 comotion via flagellar rotation notably in E. coli . A proton brane drives a water flow that turns a turbine generating energy. (b) gradient induces spontaneous transport of protons through stator Reverse electro-dialysis (RED). Fresh and sea water are separated by parts (MotA/MotB) 363. As the transport is gated through these stacks of alternating cation and anion selective membranes. Sponta- neous diffusion induces fluxes of ions through the selective membranes, channels, it induces a ratchet-like motion of the rotor part of the which is captured at the boundaries by reactive electrodes producing an 364 motor . The flagellum is attached to the rotor and therefore electric current. Usually RED is performed by alternating fresh and sea rotates. Around 1200 ions translocating per rotation generate a water a dozen times, although only three layers are represented on the force at the base of the flagellar motor of about 200 pN 361. The figure. flagellum rotates at about 100 Hz 361 allowing E. coli to swim at more than 10 body lengths per second ! The current attempts to harvest blue energy have essentially re- Nanoscale ionic machinery – The proper function of the F-type en- lied on two techniques, as sketched in Fig. 29. Pressure-retarded zyme is dependent on a subtle balance of osmotic and chemical osmosis (PRO) harvests the natural osmotic force between sea potentials for proper function 365 and the detailed mechanisms water and river water when they are separated by a semiper- involving motion and electric field coupling to the proton flux meable membrane to activate a turbine to generate electricity.

32 |Journal Name, [year], [vol.], 1–43 Reverse electro-dialysis (RED) uses diffusion gradients of salts between sea water and river water to directly generate (ionic) Iosm RL electric currents by separating the corresponding ion fluxes us- 369 ing a multi-stack of cation and anion selective membranes . electricity Iosm Both strategies rely on separation of water from ions or ions from + water, and therefore require subnanoporous structures which im- - pede the water fluxes and diminish energetic efficiency. Current + - Rpore PRO technologies are only able to produce up to 3 W/m2, less + + 2 370 - than the critical 5 W/m for economic viability . The reasons - + - for such a low performance can be readily understood: while the + + + + osmotic pressure at the interface between sea water and fresh +-+- +- - - +- -+ fresh + + + + + water is considerable and reaches 30 bars, the permeability of the advanced v water + - ∞ + Iosm semi-permeable membrane is extremely small since its pore struc- membrane + - ++ - + - ture is in the sub-nanometer scale to sieve ions: the power, which sea - - + + water λD ++ + + + + is the product of flow rate and pressure drop is accordingly small. + ++ + + On the other hand, state-of-the-art RED achieved up to 8 W/m2 ------in controlled environment 82, and there is an industrial hope for blue energy harvesting which is currently explored with the RED- Fig. 30 : Blue energy with diffusio-osmosis. A porous membrane with Stack project in the Netherlands 75,369. large and charged pores (zoom) induces a diffusio-osmotic plug-like flow with center velocity v∞ upon a salt concentration difference (e.g. here Still we note that the above power figures should not be consid- between sea and fresh water) as seen in Fig. 9). This flow drives excess ered as negligible, because membrane systems are quite compact charges in the electric double layer producing a net ionic current Iosm that and hundreds of square meters of membranes can be packed over can be harvested in a load resistance RL – top right electric- schematic. a single ground square meter. Such performances should be com- - pared to the 2.5 W per square meter of ground field required for + a Windmill farm 371, due to the very large required distance be- where Rpore is the pore or membrane resistance (that can be tween windmills to prevent flow interactions. This illustrates that obtained from standard conductance measurements). Osmotic blue energy is actually already competitive as such in spite of the power reaches thousands of Watts per square meter in BN nan- 6 2 poor performances of PRO and RED. otubes, and even up to 10 W/m for the 2D MoS2 due to its Beyond PRO and RED, it was shown recently that new nano- molecular thickness (leading to huge gradients). This estimate ac- materials and nanofluidic transport constitute key assets that al- tually suggests to couple diffusio-osmotic current generation with 75 low to boost considerably these performances 73–75,87,372. Exper- an asymmetric pore geometry leading to ionic diode behavior : iments across nanotubes of boron-nitride (BN), and subsequently blocking the ionic backflow thanks to the diode property allows one to boost the output power by reducing Joule losses (see de- across MoS2 nanoporous membranes, reported huge ionic currents. A puzzling remark is that the BN nanotubes in the ex- tails in Ref. 75 and Fig. 30). Asymmetric channels were indeed shown to improve energy harvesting 373,374. periments of Ref. 73 or the MoS2 nanopores of Ref. 74 are fully permeable to ions, in contrast to the canonical views of RED in- This methodology can be readily generalized to other materials volving cation and anion selective membranes. The origin of the which are better suited for upscaling as compared to BN nan- 75 osmotic current was then shown to be the diffusio-osmotic ionic otubes. Key progress has been made recently in this direction . currents taking place at the surface of the materials, coupled to Using diffusio-osmotic currents thus constitutes a promising route the considerable surface charge exhibited by these systems – see for improved blue energy harvesting, making it possibly relevant Fig. 30. We reported in the previous section the corresponding to industrial scale. ionic current in Eqs.(55)-(56), and for a membrane constituted Beyond these membrane-based routes, the so-called “capaci- of N tubes of radius R, length L and surface charge Σ, the ionic tive mixing” methodology is an alternative approach to harvest 375 current can be estimated as osmotic energy . The principle is to charge and discharge an ionic capacitor by alternating flows of salty and fresh water. Ca- 2πR Iosm N2πRΣ vDO N Σ DDO ∆logcs (105) pacitor plates are connected to current collectors. First (step A on ≈ × ≈ L × Fig. 31-a) salty water is flushed in, charging the capacitor plates, where vDO the diffusio-osmotic water flow speed and DDO is the resulting in a closed circuit current in the load resistance. Then diffusio-osmotic mobility, typically DDO kBT/(8πη`B). This pre- (step B), salty water is replaced by fresh water. When the circuit is ∼ diction was fully confirmed experimentally in Ref. 73. Therefore – closed again on the load resistance (step C), the capacitor plates and this is a key asset – the blue energy does not require full selec- discharge into the bulk as fresh water is less salty, resulting in a tivity of the membrane, in contrast to RED standards. Connecting current in the opposite direction. The circuit is opened and fresh the membrane to a load resistance RL –Fig. 30 –, the maximum water is replaced by salty water (step D) and the cycle may start osmotic power which can be harvested is easily found to be again. The power generated may be computed from the area of the 1 P = R I2 (106) 4 pore osm cycle in the voltage/charge plane – see Fig. 31-b. Typically, over

Journal Name, [year], [vol.],1–43 | 33 1 cycle (about 20h 376), 1 J per gram of carbon electrode may be effective on signiﬁcantly long time scales, allowing for proper re- collected. To compare with previous results, we estimate that 1 moval of the particles. carbon plate of 6 6cm2 is about 1g, such that one may recover × around 0.2 W/m2 with capacitive mixing. Capacitive mixing (a)(a)advectionadvection (b)(b) advection + diffusio-phoresis therefore requires signiﬁcant progress in optimizing the cell setup 0 min 1 min 3 min0 min5 min 1 10min min 0 min 3 min 1 min 5 min 3 10min min 0 min5 min 1 10min min 3 min 5 min 10 min and the nanoporous structure to enhance performances 377,378. x

R fresh (a)L water (b)

L + + ++ - ++ ++ + - ++ - B ++ +- + ++ - + + ++ + + - ++ - C ++ - + + - + - + - A B + - sea - - R - water L A D (c)washed and rinsed with water (d) washed with SDS , C D cellpotential (c) rinsed with water + + + + ++ + - ++ -- - ++ ++ - + - charge on electrodes porous + 1 cm1 cm 1 cm electrodes -

Fig. 31 : Capacitive Mixing to collect blue energy. (a) Capacitive mix- Fig. 32 : Particle removal with diffusio-phoresis. Reproduced from ing cycle. Two electrodes with functionalized surfaces (such that one is Ref. 379 with permission from the APS, copyright 2018. (a) Fluorescence positively charged in surface (green) and the other negatively charged in image sequence showing particles in a dead-end microfluidic pore, upon surface (orange)) are embedded in a fluidic device where salty water and advection in the main conduct. The solutions are composed of SDS at 10 fresh water are alternatively flushed in a cycle. (b) Associated Voltage mM. (b) Same as (a) with a solute gradient, where the inner pore solute versus Charge cycle. The cycle is described further in the text. concentration is 10 mM and the outer (main channel) is 0.1 mM. All scale bars are 50 µm. (c) and (d) A piece of cotton fabric is stained with colored colloidal particles (polystyrene latex). The piece of fabric is washed and rinsed in water (c) or washed in 10 mM SDS and rinsed with water (d) then photographed immediately after rinsing (left) and 120 s afterwards 6.4 Dead-end pores: Detergency, particle and liquid osmotic (right). All scale bars are 1 cm. extraction We have demonstrated in the previous sections how efficient This type of mechanism based on diffusio-phoretic migration diffusio-phoresis is to boost migration of particles. Combined is versatile and applies to any flushing geometry. Various recent with the ability to generate gradients of solute (in particular of experiments considered extraction of particles – colloidal parti- salts) at small scales, it proves a method of choice in various ap- cles and oil emulsions – from dead-end pores 209,210. Such re- plications to extract particles or liquids from dead-end pores. We sults have also obvious applications in a different context, in ge- discuss shortly two examples where diffusio-osmotic forces are ology for example, where dissolution and recrystallization at the harnessed. mineral-fluid interface leads to ubiquituous salt gradients at the A nice application of diffusio-osmosis was highlighted recently root of diffusio-phoretic and -osmotic transport 76,380. In fact, in the context of cleaning and the significance of rinsing in laun- flushing by fresh water was shown to enhance considerably oil 379 dry detergency . The question at stake here is how to extract recovery, a method coined as “Low salinity enhanced oil recov- particles which are stuck in dead-end pores in the porous matrix ery” 381. While the very origin of this phenomenon is still debated, constituting the fabric. A simple flow resulting from mechanical it is quite clear that diffusio-osmotic flows will play a key role in action may not be able to perform this task, especially since par- recovering biphasic mixtures using salinity gradients. Consider ticles buried in small pores in the interyarn pore space may not oil in a porous structure with typical pore radius a, as sketched be recovered by advection because flow is channelled by larger in Fig. 33, where oil is trapped in dead-end pores. After a flush pores (see Fig. 32-a and c). Experiments then showed that rins- with fresh water, a diffusio-osmotic flow may be generated at the ing with fresh water generates surfactant gradients at the scale of surface of the porous material, with velocity vDO = DDO∇logcs. − the fabric fibres and this in turn leads to diffusio-phoretic motion Assuming first that oil is blocked, this generates a counterbalanc- of the particles inside the dead-end pores. This flushing geom- ing pressure gradient, such that the total flux is vanishing, leading etry echoes the osmotic shock discussed above. As highlighted to a pressure drop in Fig. 32-b and d, the gradient-induced motion allows one to 8ηD extract particles from the intertwined network of pores. This sug- ∆p = DO ∆[logc ] (107) DO a2 s gests that, after laundering with any kind of detergent, rinsing − with fresh water will allow a diffusio-phoretic push to wash out along the dead-end channel (and independent of the channel dirt and stains. It is worthwhile noting that detergency within this length). Putting in numbers, with a strong salinity gradient be- prospect also benefits from the log-sensing and osmotic shock ef- tween salty water at 1 M and fresh water at 0.1 mM to fix ideas, fect discussed in Sec.5.2: particle removal by this mechanism is we find ∆pDO = 0.07 bar for a = 100nm and up to ∆pDO = 30 bars

34 |Journal Name, [year], [vol.], 1–43 for a = 5nm. This has to be compared to the oil-water capillary of processes could be improved, harvesting the properties of spe- γ pressure expressed as ∆Pcap = with γ 10 20 mN/m a typical cific geometries – as in the kidney – together with a clever mix of a ' − surface tension at the oil-water interface (possibly decorated with osmotic forces – as diffusio-phoresis for detergency. injected surfactants). While ∆pcap = 2 bar > ∆pDO for a = 100 nm, Overall we still have a lot to learn from Nature and how it it is in the same range for a = 5nm with ∆pcap ∆pDO 40 bar. harvests osmosis in many forms, for separation purposes, energy ∼ ∼ For very small pores, the pressure induced by diffusio-osmosis – storage and harvesting, etc. Today osmosis is usually harnessed which scales as 1/a2 – is thus able to bypass the capillary pressure, in its most basic form, for example as the prototypical example of scaling as 1/a. These simplistic estimates are made for illustra- osmotic pressure across a semi-permeable membrane. Yet Nature tion only and would deserve more detailed experimental investi- has developed far more clever and far more complex examples. gations. They highlight the efficiency of diffusio-osmotic effects Mimicking the natural wonders with artificial systems is a great to extract liquids which are deeply confined within nanometric challenge but it opens new avenues for many outstanding societal dead-end porosity. questions that are worth the journey. + - + - + 8 List of Symbols We report below symbols that are used frequently throughout the - + - + review. - + + - - Salinity gradient + a a Pore radius Water reservoir + - - - + Oil + - + A Membrane or pore area b Slip length of the surface β = 1/kBT cs Solute concentration cw Water or solvent concentration Fig. 33 : Diffusio-osmotic effects for oil recovery under salinity gra- c+/ Concentration of positive or negative ions − dients. Enhanced oil recovery is traditionally preformed by injecting sea DDO Diffusio-osmotic "diffusion" coefficient water in the reservoir to push the oil. However flushing with fresh water DDP Diffusio-phoretic "diffusion" coefficient slugs is known to boost the process. Gradients of salinity within dead D Colloid diffusion coefficient end pores may help bypassing the capillary forces blocking the oil within 0 the porosity. Ds Diffusion coefficient of the solute e Elementary charge E Electric field 7 Concluding remarks and perspectives ε Dielectric permittivity of the fluid As is clear from our discussion in the previous section, osmosis η Solvent viscosity is ubiquitous and crucial to an impressive number of processes, IDO Diffusio-osmotic ion current with extremely diverse manifestations. In spite of this diversity, a Ie Electric current key and universal aspect of osmosis is that it may be interpreted Je Exchange or Excess solute flow as a driving force, exerted by the membrane (or a surface, or a Js Solute flow particle’s surface, and so on) on the solute particles. As we have js = Js/A Solute flow per unit area seen in many situations in detail, we typically expect the apparent kB Boltzmann’s constant osmotic pressure to write generically as Kosm Osmotic electric mobility L Thickness of the membrane or length of the pore ∆Πapp cs( ∇U ) ' h − eff i κhyd Permeance of the membrane or pore 2 `B = e /4πεkBT Bjerrum length with . some specific average and U the effective interac- h i eff Transport matrix (or a part of the full matrix) tion potential. This mechanical perspective allows one to inter- L L = DsA , Solute permeability of the membrane or pore pret most osmotic related phenomena (diffusio-osmosis, diffusio- D ηL κhydA phoresis, active osmosis, etc.). Beyond this generic description, a Lhyd = ηL , Hydrodynamic permeability of the mem- proper description of the forces at play is required in more spe- brane or pore cific examples, as we showed on the subtle example of the force balance in diffusio-phoresis. Our understanding of osmotic related phenomena is still blurred by a number of open riddles. Non-equilibrium osmotic flows should be investigated, in particular to harvest non- equilibrium forces for advanced transport of species, which offer a number of promising avenues. Introducing more reliable de- scriptions and understanding for ionic transport at the smallest scales should also open the way to build advanced ionic detectors and ionic-powered machinery. At micrometric scales, a number

Journal Name, [year], [vol.],1–43 | 35 λ range of potential interactions 6 R. H. Dutrochet, Journal of Membrane Science, 1995, 100, λD = 1/√8π`Bcs Debye length 5–7. λ = Ds mobility of the solute s kBT 7 T. Graham et al., Philosophical Transactions of the Royal So- µDO Diffusio-osmotic mobility ciety of London, 1854, 144, 177–228. µDP Diffusio-phoretic mobility 8 A. Fick, The London, Edinburgh, and Dublin Philosophical µEO Electro-osmotic mobility Magazine and Journal of Science, 1855, 10, 30–39. 0 µi Chemical potential of the pure specie i 9 A. Mauro, Science, 1957, 126, 252–253. µi Chemical potential of specie i 10 E. Robbins and A. Mauro, The Journal of general physiology, Ni Number of molecules of specie i 1960, 43, 523–532. ω Solute "mobility" across the membrane s 11 W. Pfeffer, Osmotische untersuchungen: studien zur p Pressure zellmechanik, W. Engelmann, 1877. Π Osmotic pressure 12 G. Wald, Journal of Chemical Education, 1986, 63, 658. Q Volume flow 13 J. H. van ’t Hoff, Zeitschrift für physikalische Chemie, 1887, R Particle size 1, 481–508. ρe Charge density 14 F. G. Borg, arXiv preprint physics/0305011, 2003. σ Reflection or selectivity coefficient Σ Surface charge 15 A. Einstein, Ann. d. Phys, 1905, 17, 1. T Temperature 16 W. G. McMillan Jr and J. E. Mayer, The Journal of Chemical U (x) Potential barrier representing the membrane Physics, 1945, 13, 276–305. v Velocity field of the fluid 17 J. W. Gibbs, Nature, 1897, 55, 461. vDO Diffusio-osmotic velocity 18 H. B. Callen, Thermodynamics and an Introduction to Ther- vDP Diffusio-phoretic velocity mostatistics, 1998. vEO Electro-osmotic velocity 19 E. A. Guggenheim, Amsterdam, North-Holland, 1967, 1967, vw Molar volume of water Chapter 4. Ve Electric potential 20 J.-L. Barrat and J.-P. Hansen, Basic concepts for simple and X Solute molar fraction complex liquids, Cambridge University Press, 2003. ζ Zeta potential 21 J. S. Paustian, C. D. Angulo, R. Nery-Azevedo, N. Shi, A. I. Abdel-Fattah and T. M. Squires, Langmuir, 2015, 31, 4402– Conflicts of interest 4410. 22 J. Talen and A. Staverman, Transactions of the Faraday Soci- There are no conflicts to declare ety, 1965, 61, 2800–2804. 23 J. Talen and A. Staverman, Transactions of the Faraday Soci- Acknowledgements ety, 1965, 61, 2794–2799. The authors are grateful for the numerous discussions they en- 24 J. N. Weinstein and S. R. Caplan, Science, 1968, 161, 70–72. joyed with Marie-Laure Bocquet, Daan Frenkel, David Huang, 25 C. Lee, C. Cottin-Bizonne, A.-L. Biance, P. Joseph, L. Bocquet Jérémie Palacci, Benjamin Rotenberg, Alessandro Siria, Todd and C. Ybert, Physical Review Letters, 2014, 112, 244501. Squires, Emmanuel Trizac, Patrick Warren. Authors also acknowl- 26 C. Lee, C. Cottin-Bizonne, R. Fulcrand, L. Joly and C. Ybert, edge pertinent feedback from Maarten Biesheuvel and Alan Kay. The journal of physical chemistry letters, 2017, 8, 478–483. L.B. acknowledges funding from the European Union’s H2020 27 A. Staverman, Recueil des Travaux Chimiques des Pays-Bas, Framework Programme/ERC Advanced Grant Shadoks and Eu- 1951, 70, 344–352. ropean Union’s H2020 Framework Programme/FET NanoPhlow. 28 O. Kedem and A. Katchalsky, Biochimica et biophysica Acta, S.M. and L. B. acknowledge funding from ANR project Neptune. 1958, 27, 229–246. Notes and references 29 L. Onsager, Physical review, 1931, 37, 405. 30 S. R. De Groot and P. Mazur, Non-equilibrium thermodynam- 1 M. L. Meyer and J. van ’t Hoff, Recueil des Travaux Chimiques ics, Courier Corporation, 2013. des Pays-Bas, 1890, 9, 157–161. 31 O. Kedem and A. Katchalsky, The Journal of general physiol- 2 J. van ’t Hoff, Zeitschrift für Physikalische Chemie, 1890, 5, ogy, 1961, 45, 143–179. 174–176. 32 O. Kedem and A. Katchalsky, Transactions of the Faraday So- 3 C. Kung, B. Martinac and S. Sukharev, Annual review of mi- ciety, 1963, 59, 1918–1930. crobiology, 2010, 64, 313–329. 33 O. Kedem and A. Katchalsky, Transactions of the Faraday So- 4 D. C. Guell and H. Brenner, Industrial & engineering chem- ciety, 1963, 59, 1931–1940. istry research, 1996, 35, 3004–3014. 34 O. Kedem and A. Katchalsky, Transactions of the Faraday So- 5 R. J. H. Dutrochet, L’agent immédiat du mouvement vital ciety, 1963, 59, 1941–1953. dévoilé dans sa nature et dans son mode d’action, chez les veg- 35 E. H. Starling, The Journal of physiology, 1896, 19, 312–326. étaux et chez les animaux, Baillière, 1826.

36 |Journal Name, [year], [vol.], 1–43 36 J. R. Pappenheimer, Physiological Reviews, 1953, 33, 387– 64 T. Mouterde and L. Bocquet, The European physical journal 423. E, 2018, 41, 148. 37 R. Adamson, J. Lenz, X. Zhang, G. Adamson, S. Weinbaum 65 R. Messinger and T. Squires, Physical Review Letters, 2010, and F. Curry, The Journal of physiology, 2004, 557, 889–907. 105, 144503. 38 J. L. Anderson and D. M. Malone, Biophysical journal, 1974, 66 A. J. Pascall and T. M. Squires, Physical Review Letters, 2010, 14, 957–982. 104, 088301. 39 A. Yamauchi, Y. Shin, M. Shinozaki and M. Kawabe, Journal 67 D. J. Bonthuis and R. R. Netz, Langmuir, 2012, 28, 16049– of Membrane Science, 2000, 170, 1–7. 16059. 40 H. Fujita and Y. Kobatake, Journal of Colloid and Interface 68 J. L. Anderson and D. C. Prieve, Langmuir, 1991, 7, 403– Science, 1968, 27, 609–615. 406. 41 A. Hill, Quarterly reviews of biophysics, 1979, 12, 67–99. 69 A. Ajdari and L. Bocquet, Physical Review Letters, 2006, 96, 42 J. D. Ferry, Chemical Reviews, 1936, 18, 373–455. 186102. 43 P. M. Ray, Plant physiology, 1960, 35, 783. 70 D. M. Huang, C. Cottin-Bizonne, C. Ybert and L. Bocquet, 44 J. C. Giddings, E. Kucera, C. P. Russell and M. N. Myers, The Physical Review Letters, 2008, 101, 064503. Journal of Physical Chemistry, 1968, 72, 4397–4408. 71 D. Prieve, J. Anderson, J. Ebel and M. Lowell, Journal of 45 E. M. Renkin, The Journal of general physiology, 1954, 38, Fluid Mechanics, 1984, 148, 247–269. 225–243. 72 M. Lokesh, S. K. Youn and H. G. Park, Nano letters, 2018, 18, 46 J. L. Anderson and J. A. Quinn, Biophysical Journal, 1974, 6679–6685. 14, 130. 73 A. Siria, P. Poncharal, A.-L. Biance, R. Fulcrand, X. Blase, 47 T. Chou, Physical Review Letters, 1998, 80, 85. S. T. Purcell and L. Bocquet, Nature, 2013, 494, 455. 48 T. Chou, The Journal of Chemical Physics, 1999, 110, 606– 74 J. Feng, M. Graf, K. Liu, D. Ovchinnikov, D. Dumcenco, 615. M. Heiranian, V. Nandigana, N. R. Aluru, A. Kis and A. Rade- 49 G. S. Manning, The Journal of Chemical Physics, 1968, 49, novic, Nature, 2016, 536, 197. 2668–2675. 75 A. Siria, M.-L. Bocquet and L. Bocquet, Nature Reviews Chem- 50 C. B. Picallo, S. Gravelle, L. Joly, E. Charlaix and L. Bocquet, istry, 2017, 1, 0091. Physical Review Letters, 2013, 111, 244501. 76 O. Plümper, A. Botan, C. Los, Y. Liu, A. Malthe-Sørenssen 51 S. Marbach, H. Yoshida and L. Bocquet, The Journal of Chem- and B. Jamtveit, Nature geoscience, 2017, 10, 685. ical Physics, 2017, 146, 194701. 77 J. Palacci, C. Cottin-Bizonne, C. Ybert and L. Bocquet, Soft 52 P. J. W. Debye, P. Debye, B. Eckstein, W. Barber and G. Ar- Matter, 2012, 8, 980–994. quette, Equilibrium and sedimentation of uncharged particles 78 J. Palacci, B. Abécassis, C. Cottin-Bizonne, C. Ybert and in inhomogeneous electric fields, Academic Press, 1954, pp. L. Bocquet, Physical Review Letters, 2010, 104, 138302. 273–285. 79 M. Kosmulski and E. Matuevi, Journal of colloid and interface 53 E. Grim and K. Sollner, The Journal of general physiology, science, 1992, 150, 291–294. 1957, 40, 887–899. 80 N. Shi, R. Nery-Azevedo, A. I. Abdel-Fattah and T. M. 54 J. L. Anderson, Annual review of fluid mechanics, 1989, 21, Squires, Physical Review Letters, 2016, 117, 258001. 61–99. 81 J. Fair and J. Osterle, The Journal of Chemical Physics, 1971, 55 B. Derjaguin, G. Sidorenkov, E. Zubashchenko and E. Kisel- 54, 3307–3316. eva, Progress in Surface Science, 1993, 43, 138–152. 82 D.-K. Kim, C. Duan, Y.-F. Chen and A. Majumdar, Microflu- 56 B. Derjaguin, S. Dukhin and A. Korotkova, Progress in Surface idics and Nanofluidics, 2010, 9, 1215–1224. Science, 1993, 43, 153–158. 83 Z. Zhang, X. Sui, P. Li, G. Xie, X.-Y. Kong, K. Xiao, L. Gao, 57 R. J. Hunter, Foundations of colloid science, Oxford university L. Wen and L. Jiang, Journal of the American Chemical Soci- press, 2001. ety, 2017, 139, 8905–8914. 58 D. Andelman, Handbook of biological physics, Elsevier, 1995, 84 M. Wanunu, W. Morrison, Y. Rabin, A. Y. Grosberg and vol. 1, pp. 603–642. A. Meller, Nature nanotechnology, 2010, 5, 160. 59 T. M. Squires, Fluids, Colloids and Soft Materials: An Intro- 85 Z. S. Siwy, Advanced Functional Materials, 2006, 16, 735– duction to Soft Matter Physics, 2016, 59–79. 746. 60 R. B. Schoch, J. Han and P. Renaud, Reviews of modern 86 N. A. Bell, C. R. Engst, M. Ablay, G. Divitini, C. Ducati, physics, 2008, 80, 839. T. Liedl and U. F. Keyser, Nano letters, 2011, 12, 512–517. 61 S. Devasenathipathy and J. Santiago, Microscale Diagnostic 87 M. I. Walker, K. Ubych, V. Saraswat, E. A. Chalklen, Techniques, Springer, 2005, pp. 113–154. P. Braeuninger-Weimer, S. Caneva, R. S. Weatherup, S. Hof- 62 L. Bocquet and E. Charlaix, Chemical Society Reviews, 2010, mann and U. F. Keyser, ACS nano, 2017, 11, 1340–1346. 39, 1073–1095. 88 L. Wang, M. S. Boutilier, P. R. Kidambi, D. Jang, N. G. Had- 63 A. S. Khair and T. M. Squires, Physics of Fluids, 2009, 21, jiconstantinou and R. Karnik, Nature nanotechnology, 2017, 042001.

Journal Name, [year], [vol.],1–43 | 37 12, 509. S. Duhr, Nature communications, 2010, 1, 1–7. 89 J. E. Hall, The Journal of general physiology, 1975, 66, 531– 117 T. Dau, E. Edeleva, S. Seidel, R. Stockley, D. Braun and D. E. 532. Jenne, Scientiﬁc reports, 2016, 6, 35413. 90 M. Mao, J. Sherwood and S. Ghosal, Journal of Fluid Me- 118 A. F. Al-Alawy and R. M. Al-Alawy, Iraqi Journal of Chemical chanics, 2014, 749, 167–183. and Petroleum Engineering, 2016, 17, 53–68. 91 J. D. Sherwood, M. Mao and S. Ghosal, Langmuir, 2014, 30, 119 N. Kuipers, J. H. Hanemaaijer, H. Brouwer, J. van Mede- 9261–9272. voort, A. Jansen, F. Altena, P. van der Vleuten and H. Bak, 92 D. V. Melnikov, Z. K. Hulings and M. E. Gracheva, Physical Desalination and Water Treatment, 2015, 55, 2766–2776. Review E, 2017, 95, 063105. 120 A. P. Straub, N. Y. Yip, S. Lin, J. Lee and M. Elimelech, Nature 93 D. J. Rankin, L. Bocquet and D. M. Huang, submitted, 2019, Energy, 2016, 1, 16090. arxiv.org/abs/1904.10636. 121 B. Rotenberg and I. Pagonabarraga, Molecular Physics, 2013, 94 R. A. Sampson, Philosophical Transactions of the Royal Society 111, 827–842. of London. A, 1891, 182, 449–518. 122 A. Kalra, S. Garde and G. Hummer, Proceedings of the Na- 95 Z. Dagan, S. Weinbaum and R. Pfeffer, Journal of Fluid Me- tional Academy of Sciences, 2003, 100, 10175–10180. chanics, 1982, 115, 505–523. 123 Y. Luo and B. Roux, The Journal of Physical Chemistry Letters, 96 J. Happel and H. Brenner, Low Reynolds number hydro- 2009, 1, 183–189. dynamics: with special applications to particulate media, 124 T. W. Lion and R. J. Allen, The Journal of Chemical Physics, Springer Science & Business Media, 2012, vol. 1. 2012, 137, 244911. 97 G. Lippmann, Compt. rend, 1907, 145, 104–105. 125 H. Yoshida, S. Marbach and L. Bocquet, The Journal of Chem- 98 M. Aubert, Ann. Chim. Phys., 1912, 26, 165. ical Physics, 2017, 146, 194702. 99 B. V. Derjaguin, Surface Forces and Surfactant Systems, 126 I. McDonald and J. Hansen, Academic Press, London, ed, Springer, 1987, pp. 17–30. 1986, 2, 179. 100 B. Derjaguin, N. Churaev and V. Muller, Surface Forces, 127 V. Marry, J.-F. Dufrêche, M. Jardat and P. Turq, Molecular Springer, 1987, pp. 390–409. Physics, 2003, 101, 3111–3119. 101 A. P. Bregulla, A. Würger, K. Günther, M. Mertig and F. Ci- 128 L. Bocquet and J.-L. Barrat, Physical review E, 1994, 49, chos, Physical Review Letters, 2016, 116, 188303. 3079. 102 V. M. Barragán and S. Kjelstrup, Journal of Non-Equilibrium 129 H. Yoshida, H. Mizuno, T. Kinjo, H. Washizu and J.-L. Barrat, Thermodynamics, 2017, 42, 217–236. Physical Review E, 2014, 90, 052113. 103 E. Ruckenstein, Journal of Colloid and Interface Science, 130 H. Yoshida, H. Mizuno, T. Kinjo, H. Washizu and J.-L. Barrat, 1981, 83, 77–81. The Journal of chemical physics, 2014, 140, 214701. 104 R. Piazza, Journal of Physics: Condensed Matter, 2004, 16, 131 Y. Liu, R. Ganti, H. G. Burton, X. Zhang, W. Wang and S4195. D. Frenkel, Physical review letters, 2017, 119, 224502. 105 A. Parola and R. Piazza, The European Physical Journal E, 132 Y. Liu, R. Ganti and D. Frenkel, Journal of Physics: Condensed 2004, 15, 255–263. Matter, 2018, 30, 205002. 106 S. Duhr and D. Braun, Proceedings of the National Academy 133 R. Ganti, Y. Liu and D. Frenkel, Physical Review Letters, 2017, of Sciences, 2006, 103, 19678–19682. 119, 038002. 107 R. Piazza, Soft Matter, 2008, 4, 1740–1744. 134 R. Ganti, Y. Liu and D. Frenkel, Physical Review Letters, 2018, 108 R. Piazza and A. Parola, Journal of Physics: Condensed Mat- 121, 068002. ter, 2008, 20, 153102. 135 K. Kiyosawa and M. Tazawa, Protoplasma, 1973, 78, 203– 109 J. Morthomas and A. Würger, Journal of Physics: Condensed 214. Matter, 2008, 21, 035103. 136 D. J. Bonthuis and R. Golestanian, Physical Review Letters, 110 L. Fu, S. Merabia and L. Joly, Physical Review Letters, 2017, 2014, 113, 148101. 119, 214501. 137 A. Esfandiar, B. Radha, F. Wang, Q. Yang, S. Hu, S. Garaj, 111 R. Golestanian, T. Liverpool and A. Ajdari, New Journal of R. Nair, A. Geim and K. Gopinadhan, Science, 2017, 358, Physics, 2007, 9, 126. 511–513. 112 R. Di Leonardo, F. Ianni and G. Ruocco, Langmuir, 2009, 25, 138 W. Choi, Z. W. Ulissi, S. F. Shimizu, D. O. Bellisario, M. D. 4247–4250. Ellison and M. S. Strano, Nature communications, 2013, 4, 113 A. Würger, Reports on Progress in Physics, 2010, 73, 126601. 2397. 114 L.-H. Yu and Y.-F. Chen, Analytical chemistry, 2015, 87, 139 E. Secchi, S. Marbach, A. Niguès, D. Stein, A. Siria and 2845–2851. L. Bocquet, Nature, 2016, 537, 210. 115 H.-R. Jiang, H. Wada, N. Yoshinaga and M. Sano, Physical 140 R. H. Tunuguntla, R. Y. Henley, Y.-C. Yao, T. A. Pham, M. Wa- Review Letters, 2009, 102, 208301. nunu and A. Noy, Science, 2017, 357, 792–796. 116 C. J. Wienken, P. Baaske, U. Rothbauer, D. Braun and 141 K. A. Mahmoud, B. Mansoor, A. Mansour and M. Khraisheh, Desalination, 2015, 356, 208–225.

38 |Journal Name, [year], [vol.], 1–43 142 R. Nair, H. Wu, P. Jayaram, I. Grigorieva and A. Geim, Sci- 178. ence, 2012, 335, 442–444. 167 J. Rousselet, L. Salome, A. Ajdari and J. Prostt, Nature, 1994, 143 B. Radha, A. Esfandiar, F. Wang, A. Rooney, K. Gopinadhan, 370, 446. A. Keerthi, A. Mishchenko, A. Janardanan, P. Blake, L. Fu- 168 S. Marbach, N. Kavokine and L. Bocquet, submitted to J. magalli et al., Nature, 2016, 538, 222. Chem. Phys., 2019. 144 F. Fornasiero, H. G. Park, J. K. Holt, M. Stadermann, C. P. 169 R. Karnik, R. Fan, M. Yue, D. Li, P. Yang and A. Majumdar, Grigoropoulos, A. Noy and O. Bakajin, Proceedings of the Na- Nano letters, 2005, 5, 943–948. tional Academy of Sciences, 2008, 105, 17250–17255. 170 E. B. Kalman, O. Sudre, I. Vlassiouk and Z. S. Siwy, Analytical 145 J. Feng, K. Liu, M. Graf, D. Dumcenco, A. Kis, M. Di Ventra and bioanalytical chemistry, 2009, 394, 413–419. and A. Radenovic, Nature materials, 2016, 15, 850. 171 W. Guan, R. Fan and M. A. Reed, Nature communications, 146 M. Majumder, N. Chopra, R. Andrews and B. J. Hinds, Na- 2011, 2, 506. ture, 2005, 438, 44. 172 K.-G. Zhou, K. Vasu, C. Cherian, M. Neek-Amal, J. C. 147 J. K. Holt, H. G. Park, Y. Wang, M. Stadermann, A. B. Zhang, H. Ghorbanfekr-Kalashami, K. Huang, O. Mar- Artyukhin, C. P. Grigoropoulos, A. Noy and O. Bakajin, Sci- shall, V. Kravets, J. Abraham et al., arXiv preprint ence, 2006, 312, 1034–1037. arXiv:1805.06390, 2018. 148 M. Whitby, L. Cagnon, M. Thanou and N. Quirke, Nano let- 173 J. Moorthy, C. Khoury, J. S. Moore and D. J. Beebe, Sensors ters, 2008, 8, 2632–2637. and Actuators B: Chemical, 2001, 75, 223–229. 149 J. Lee, T. Laoui and R. Karnik, Nature nanotechnology, 2014, 174 H. Zhang, X. Hou, L. Zeng, F. Yang, L. Li, D. Yan, Y. Tian 9, 317. and L. Jiang, Journal of the American Chemical Society, 2013, 150 M. Langecker, V. Arnaut, T. G. Martin, J. List, S. Renner, 135, 16102–16110. M. Mayer, H. Dietz and F. C. Simmel, Science, 2012, 338, 175 P. Liu, G. Xie, P. Li, Z. Zhang, L. Yang, Y. Zhao, C. Zhu, X.-Y. 932–936. Kong, L. Jiang and L. Wen, NPG Asia Materials, 2018, 10, 151 R. Joshi, P. Carbone, F. C. Wang, V. G. Kravets, Y. Su, I. V. 849–857. Grigorieva, H. Wu, A. K. Geim and R. R. Nair, science, 2014, 176 J. Liu, N. Wang, L.-J. Yu, A. Karton, W. Li, W. Zhang, F. Guo, 343, 752–754. L. Hou, Q. Cheng, L. Jiang et al., Nature communications, 152 M. Majumder, A. Siria and L. Bocquet, MRS Bulletin, 2017, 2017, 8, 2011. 42, 278–282. 177 L. Hu, S. Gao, X. Ding, D. Wang, J. Jiang, J. Jin and L. Jiang, 153 R. Karnik, C. Duan, K. Castelino, H. Daiguji and A. Majum- ACS nano, 2015, 9, 4835–4842. dar, Nano letters, 2007, 7, 547–551. 178 D. C. Prieve and R. Roman, Journal of the Chemical Soci- 154 A. Poggioli, A. Siria and L. Bocquet, The journal of physical ety, Faraday Transactions 2: Molecular and Chemical Physics, chemistry B, 2019, 123, 1171–1185. 1987, 83, 1287–1306. 155 R. E. Farmer and R. I. Macey, Biochimica et Biophysica Acta 179 J. Ebel, J. L. Anderson and D. Prieve, Langmuir, 1988, 4, (BBA)-Biomembranes, 1970, 196, 53–65. 396–406. 156 C. Toupin, M. Le Maguer and L. McGann, Cryobiology, 1989, 180 D. Velegol, A. Garg, R. Guha, A. Kar and M. Kumar, Soft 26, 431–444. Matter, 2016, 12, 4686–4703. 157 D. B. Peckys, F. Kleinhans and P. Mazur, PloS one, 2011, 6, 181 F. M. Möller, F. Kriegel, M. Kieß, V. Sojo and D. Braun, Ange- e23643. wandte Chemie International Edition, 2017, 56, 2340–2344. 158 S. Marbach and L. Bocquet, The Journal of Chemical Physics, 182 A. Banerjee, I. Williams, R. N. Azevedo, M. E. Helgeson and 2017, 147, 154701. T. M. Squires, Proceedings of the National Academy of Sci- 159 S. Marbach, D. S. Dean and L. Bocquet, Nature Physics, 2018, ences, 2016, 113, 8612–8617. 14, 1108–1113. 183 F. Morrison Jr, Journal of Colloid and Interface Science, 1970, 160 Y. Sakiyama, A. Mazur, L. E. Kapinos and R. Y. Lim, Nature 34, 210–214. nanotechnology, 2016, 11, 719. 184 B. Rallabandi, F. Yang and H. A. Stone, arXiv preprint 161 S. Y. Noskov, S. Berneche and B. Roux, Nature, 2004, 431, arXiv:1901.04311, 2019. 830. 185 A. Chamolly, T. Ishikawa and E. Lauga, New Journal of 162 G. Eisenman and R. Horn, The Journal of membrane biology, Physics, 2017, 19, 115001. 1983, 76, 197–225. 186 U. M. Córdova-Figueroa and J. F. Brady, Physical Review Let- 163 P. Läuger, W. Stephan and E. Frehland, Biochimica et Bio- ters, 2008, 100, 158303. physica Acta (BBA)-Biomembranes, 1980, 602, 167–180. 187 F. Jülicher and J. Prost, Physical Review Letters, 2009, 103, 164 H. Schröder, The Journal of chemical physics, 1983, 79, 079801. 1997–2005. 188 T. M. Fischer and P. Dhar, Physical Review Letters, 2009, 102, 165 L. Gammaitoni, P. Hänggi, P. Jung and F. Marchesoni, Re- 159801. views of modern physics, 1998, 70, 223. 189 U. M. Córdova-Figueroa and J. F. Brady, Phys. Rev. Lett., 166 P. Reimann and P. Hänggi, Applied Physics A, 2002, 75, 169– 2009, 103, 079802.

Journal Name, [year], [vol.],1–43 | 39 190 U. M. Córdova-Figueroa and J. F. Brady, Phys. Rev. Lett., ture, 2010, 467, 692. 2009, 102, 159802. 216 C. J. Myers, M. Celebrano and M. Krishnan, Nature nanotech- 191 J. F. Brady, Journal of Fluid Mechanics, 2011, 667, 216–259. nology, 2015, 10, 886. 192 J. L. Moran and J. D. Posner, Annual Review of Fluid Mechan- 217 J. Agudo-Canalejo, T. Adeleke-Larodo, P. Illien and ics, 2017, 49, 511–540. R. Golestanian, Accounts of chemical research, 2018, 51, 193 B. Sabass and U. Seifert, The Journal of chemical physics, 2365–2372. 2012, 136, 064508. 218 J. Lutkenhaus, Science, 2008, 320, 755–756. 194 N. Shariﬁ-Mood, J. Koplik and C. Maldarelli, Physics of Flu- 219 L. Rothﬁeld, A. Taghbalout and Y.-L. Shih, Nature Reviews ids, 2013, 25, 012001. Microbiology, 2005, 3, 959. 195 U. Córdova-Figueroa, J. Brady and S. Shklyaev, Soft Matter, 220 A. Eldar, R. Dorfman, D. Weiss, H. Ashe, B.-Z. Shilo and 2013, 9, 6382–6390. N. Barkai, Nature, 2002, 419, 304. 196 H. Ohshima, T. W. Healy and L. R. White, Journal of the 221 R. Sear, 2019, ArXiv:1901.00802,. Chemical Society, Faraday Transactions 2: Molecular and 222 A. Aubret, S. Ramananarivo and J. Palacci, Current Opinion Chemical Physics, 1983, 79, 1613–1628. in Colloid & Interface Science, 2017, 30, 81–89. 197 D. Long, J.-L. Viovy and A. Ajdari, Physical Review Letters, 223 P. Illien, R. Golestanian and A. Sen, Chemical Society Reviews, 1996, 76, 3858. 2017, 46, 5508–5518. 198 H.-R. Jiang and M. Sano, Applied Physics Letters, 2007, 91, 224 C. Bechinger, R. Di Leonardo, H. Löwen, C. Reichhardt, 154104. G. Volpe and G. Volpe, Reviews of Modern Physics, 2016, 88, 199 S. Michelin and E. Lauga, Journal of Fluid Mechanics, 2014, 045006. 747, 572–604. 225 P. H. Colberg, S. Y. Reigh, B. Robertson and R. Kapral, Ac- 200 B. Abécassis, C. Cottin-Bizonne, C. Ybert, A. Ajdari and counts of chemical research, 2014, 47, 3504–3511. L. Bocquet, Nature materials, 2008, 7, 785. 226 W. F. Paxton, K. C. Kistler, C. C. Olmeda, A. Sen, S. K. St. An- 201 Y. V. Kalinin, L. Jiang, Y. Tu and M. Wu, Biophysical journal, gelo, Y. Cao, T. E. Mallouk, P. E. Lammert and V. H. Crespi, 2009, 96, 2439–2448. Journal of the American Chemical Society, 2004, 126, 13424– 202 R. P. Sear and P. B. Warren, Physical Review E, 2017, 96, 13431. 062602. 227 N. Mano and A. Heller, Journal of the American Chemical 203 J. S. Paustian, R. N. Azevedo, S.-T. B. Lundin, M. J. Gilkey Society, 2005, 127, 11574–11575. and T. M. Squires, Physical Review X, 2013, 3, 041010. 228 J. R. Howse, R. A. Jones, A. J. Ryan, T. Gough, R. Vafabakhsh 204 R. Nery-Azevedo, A. Banerjee and T. M. Squires, Langmuir, and R. Golestanian, Physical Review Letters, 2007, 99, 2017, 33, 9694–9702. 048102. 205 S. Shin, O. Shardt, P. B. Warren and H. A. Stone, Nature 229 J. Palacci, C. Cottin-Bizonne, C. Ybert and L. Bocquet, Physi- Communications, 2017, 8, 15181. cal Review Letters, 2010, 105, 088304. 206 R. Guha, F. Mohajerani, M. Collins, S. Ghosh, A. Sen and 230 I. Theurkauff, C. Cottin-Bizonne, J. Palacci, C. Ybert and D. Velegol, Journal of the American Chemical Society, 2017, L. Bocquet, Physical Review Letters, 2012, 108, 268303. 139, 15588–15591. 231 J. Palacci, S. Sacanna, A. P. Steinberg, D. J. Pine and P. M. 207 D. Florea, S. Musa, J. M. Huyghe and H. M. Wyss, Proceed- Chaikin, Science, 2013, 1230020. ings of the National Academy of Sciences, 2014, 201322857. 232 I. Buttinoni, J. Bialké, F. Kümmel, H. Löwen, C. Bechinger 208 J. Nardi, R. Bruinsma and E. Sackmann, Physical Review Let- and T. Speck, Physical Review Letters, 2013, 110, 238301. ters, 1999, 82, 5168. 233 G. Mino, T. E. Mallouk, T. Darnige, M. Hoyos, J. Dauchet, 209 A. Kar, T.-Y. Chiang, I. Ortiz Rivera, A. Sen and D. Velegol, J. Dunstan, R. Soto, Y. Wang, A. Rousselet and E. Clement, ACS nano, 2015, 9, 746–753. Physical Review Letters, 2011, 106, 048102. 210 S. Shin, E. Um, B. Sabass, J. T. Ault, M. Rahimi, P. B. War- 234 F. Ginot, I. Theurkauff, D. Levis, C. Ybert, L. Bocquet, ren and H. A. Stone, Proceedings of the National Academy of L. Berthier and C. Cottin-Bizonne, Physical Review X, 2015, Sciences, 2016, 113, 257–261. 5, 011004. 211 S. Shin, J. T. Ault, P. B. Warren and H. A. Stone, Physical 235 A. Aubret, M. Youssef, S. Sacanna and J. Palacci, Nature Review X, 2017, 7, 041038. Physics, 2018, 14, 1114–1118. 212 L. Shapiro, H. H. McAdams and R. Losick, Science, 2009, 236 A. Brown and W. Poon, Soft matter, 2014, 10, 4016–4027. 326, 1225–1228. 237 A. T. Brown, W. C. Poon, C. Holm and J. de Graaf, Soft Mat- 213 M. Osawa, D. E. Anderson and H. P. Erickson, Science, 2008, ter, 2017, 13, 1200–1222. 320, 792–794. 238 P. Lammert, J. Prost and R. Bruinsma, Journal of theoretical 214 M. Loose, E. Fischer-Friedrich, J. Ries, K. Kruse and biology, 1996, 178, 387–391. P. Schwille, Science, 2008, 320, 789–792. 239 R. Golestanian, T. B. Liverpool and A. Ajdari, Physical Review 215 M. Krishnan, N. Mojarad, P. Kukura and V. Sandoghdar, Na- Letters, 2005, 94, 220801. 240 R. Di Leonardo, L. Angelani, D. Dell’Arciprete, G. Ruocco,

40 |Journal Name, [year], [vol.], 1–43 V. Iebba, S. Schippa, M. Conte, F. Mecarini, F. De Angelis 2011, 7, 181–186. and E. Di Fabrizio, Proceedings of the National Academy of 266 V. C. Onishi, A. Carrero-Parreno, J. A. Reyes-Labarta, E. S. Sciences, 2010, 107, 9541–9545. Fraga and J. A. Caballero, Journal of Cleaner Production, 241 T. W. Lion and R. J. Allen, EPL (Europhysics Letters), 2014, 2017, 140, 1399–1414. 106, 34003. 267 A. Y. Hoekstra and M. M. Mekonnen, Proceedings of the na- 242 A. P. Solon, Y. Fily, A. Baskaran, M. E. Cates, Y. Kafri, M. Kar- tional academy of sciences, 2012, 109, 3232–3237. dar and J. Tailleur, Nature Physics, 2015, 11, 673. 268 A. Alkaisi, R. Mossad and A. Shariﬁan-Barforoush, Energy 243 J. Rodenburg, M. Dijkstra and R. van Roij, Soft matter, 2017, Procedia, 2017, 110, 268–274. 13, 8957–8963. 269 G. Amy, N. Ghaffour, Z. Li, L. Francis, R. V. Linares, T. Mis- 244 S. C. Takatori, R. De Dier, J. Vermant and J. F. Brady, Nature simer and S. Lattemann, Desalination, 2017, 401, 16–21. communications, 2016, 7, 10694. 270 J. R. McCutcheon, R. L. McGinnis and M. Elimelech, Desali- 245 J. Tailleur and M. Cates, Physical Review Letters, 2008, 100, nation, 2005, 174, 1–11. 218103. 271 R. V. Linares, Z. Li, V. Yangali-Quintanilla, N. Ghaffour, 246 P. G. Moerman, H. W. Moyses, E. B. Van Der Wee, D. G. Grier, G. Amy, T. Leiknes and J. S. Vrouwenvelder, Water research, A. Van Blaaderen, W. K. Kegel, J. Groenewold and J. Brujic, 2016, 88, 225–234. Physical Review E, 2017, 96, 032607. 272 Q. Chen, Q. Ge, W. Xu and W. Pan, Journal of Membrane 247 S. Y. Reigh, P. Chuphal, S. Thakur and R. Kapral, Soft matter, Science, 2019, 574, 10–16. 2018, 14, 6043–6057. 273 T. D. Wheeler and A. D. Stroock, Nature, 2008, 455, 208. 248 F. Ginot, I. Theurkauff, F. Detcheverry, C. Ybert and 274 R. Semiat, Environmental science & technology, 2008, 42, C. Cottin-Bizonne, Nature communications, 2018, 9, 696. 8193–8201. 249 E. F. Keller and L. A. Segel, Journal of theoretical biology, 275 Y.-M. Chao and T. Liang, Desalination, 2008, 221, 433–439. 1970, 26, 399–415. 276 Y. Oren, Desalination, 2008, 228, 10–29. 250 M. P. Brenner, L. S. Levitov and E. O. Budrene, Biophysical 277 M. E. Suss, T. F. Baumann, W. L. Bourcier, C. M. Spadac- journal, 1998, 74, 1677–1693. cini, K. A. Rose, J. G. Santiago and M. Stadermann, Energy 251 R. Golestanian, Physical Review Letters, 2012, 108, 038303. & Environmental Science, 2012, 5, 9511–9519. 252 Y. Zhang, A. McMullen, L.-L. Pontani, X. He, R. Sha, N. C. 278 T. Kim, C. A. Gorski and B. E. Logan, Environmental Science Seeman, J. Brujic and P. M. Chaikin, Nature Communica- & Technology Letters, 2017, 4, 444–449. tions, 2017, 8, 21. 279 D. Deng, E. V. Dydek, J.-H. Han, S. Schlumpberger, A. Mani, 253 O. Pohl and H. Stark, Physical Review Letters, 2014, 112, B. Zaltzman and M. Z. Bazant, Langmuir, 2013, 29, 16167– 238303. 16177. 254 H. Stark, Accounts of chemical research, 2018, 51, 2681– 280 S. J. Kim, S. H. Ko, K. H. Kang and J. Han, Nature Nanotech- 2688. nology, 2010, 5, 297. 255 L. Feng, R. Dreyfus, R. Sha, N. C. Seeman and P. M. Chaikin, 281 M. W. Shahzad, M. Burhan, L. Ang and K. C. Ng, Emerging Advanced Materials, 2013, 25, 2779–2783. Technologies for Sustainable Desalination Handbook, Elsevier, 256 A. McMullen, M. Holmes-Cerfon, F. Sciortino, A. Y. Grosberg 2018, pp. 3–34. and J. Brujic, Physical Review Letters, 2018, 121, 138002. 282 I. V. Arámburo-Miranda and E. H. Ruelas-Ramírez, Environ- 257 P. Sheeler and D. E. Bianchi, Cell and molecular biology, Wi- mental Science and Pollution Research, 2017, 24, 25676– ley New York etc., 1987. 25681. 258 K. H. Jensen, E. Rio, R. Hansen, C. Clanet and T. Bohr, Jour- 283 K. Minas, E. Karunakaran, T. Bond, C. Gandy, A. Hons- nal of Fluid Mechanics, 2009, 636, 371–396. bein, M. Madsen, J. Amezaga, A. Amtmann, M. Templeton, 259 J. Comtet, K. H. Jensen, R. Turgeon, A. D. Stroock and C. Biggs et al., Desalination and Water Treatment, 2015, 55, A. Hosoi, Nature plants, 2017, 3, 17032. 2647–2668. 260 B. E. Logan and M. Elimelech, Nature, 2012, 488, 313. 284 H. Liu, R. Ramnarayanan and B. E. Logan, Environmental 261 J. R. Werber, C. O. Osuji and M. Elimelech, Nature Reviews science & technology, 2004, 38, 2281–2285. Materials, 2016, 1, 16018. 285 M. Elimelech and W. A. Phillip, science, 2011, 333, 712–717. 262 E. Jones, M. Qadir, M. T. van Vliet, V. Smakhtin and S.-M. 286 A. Mehta and A. L. Zydney, Journal of Membrane Science, Kang, Science of the Total Environment, 2019, 657, 1343– 2005, 249, 245–249. 1356. 287 H. B. Park, J. Kamcev, L. M. Robeson, M. Elimelech and B. D. 263 W. H. Organization, W. J. W. Supply and S. M. Programme, Freeman, Science, 2017, 356, eaab0530. Progress on sanitation and drinking water: 2015 update and 288 J. R. Werber, A. Deshmukh and M. Elimelech, Environmental MDG assessment, World Health Organization, 2015. Science & Technology Letters, 2016, 3, 112–120. 264 R. D. Vidic, S. L. Brantley, J. M. Vandenbossche, D. Yox- 289 L. W. Stephens, R. Lindsay, M. Milne, A. Klein, V. Fuchs and theimer and J. D. Abad, Science, 2013, 340, 1235009. L. Rodenburg, Proceedings of the Water Environment Federa- 265 K. B. Gregory, R. D. Vidic and D. A. Dzombak, Elements, tion, 2016, 2016, 5423–5429.

Journal Name, [year], [vol.],1–43 | 41 290 L. Zeman and M. Wales, Synthetic membranes, 1981, 2, 411– 315 K. Meyer, O. Ostrenko, G. Bourantas, H. Morales-Navarrete, 434. N. Porat-Shliom, F. Segovia-Miranda, H. Nonaka, A. Ghaemi, 291 G. M. Geise, H. B. Park, A. C. Sagle, B. D. Freeman and J. E. J.-M. Verbavatz, L. Brusch et al., Cell systems, 2017, 4, 277– McGrath, Journal of Membrane Science, 2011, 369, 130–138. 290. 292 M. Barboiu, Chemical Communications, 2016, 52, 5657– 316 J. R. Thiagarajah and A. S. Verkman, Physiology of the Gas- 5665. trointestinal Tract (Sixth Edition), Elsevier, 2018, pp. 1249– 293 Y.-x. Shen, W. C. Song, D. R. Barden, T. Ren, C. Lang, 1272. H. Feroz, C. B. Henderson, P. O. Saboe, D. Tsai, H. Yan et al., 317 R. J. Turner and H. Sugiya, Oral diseases, 2002, 8, 3–11. Nature communications, 2018, 9, 2294. 318 E. Gin, E. M. Tanaka and L. Brusch, Journal of theoretical 294 Z. Sun, I. Kocsis, Y. Li, Y.-M. Legrand and M. Barboiu, Fara- biology, 2010, 264, 1077–1088. day discussions, 2018, 209, 113–124. 319 C. Rauch and E. Farge, Biophysical journal, 2000, 78, 3036– 295 M. S. Mauter, I. Zucker, F. Perreault, J. R. Werber, J.-H. Kim 3047. and M. Elimelech, Nature Sustainability, 2018, 1, 166. 320 A. Evilevitch, L. Lavelle, C. M. Knobler, E. Raspaud and 296 P. Nednoor, V. G. Gavalas, N. Chopra, B. J. Hinds and L. G. W. M. Gelbart, Proceedings of the National Academy of Sci- Bachas, Journal of Materials Chemistry, 2007, 17, 1755– ences, 2003, 100, 9292–9295. 1757. 321 J. Hubbard, S. Jones and E. Landau, The Journal of physiol- 297 W.-F. Chan, E. Marand and S. M. Martin, Journal of Mem- ogy, 1968, 197, 639–657. brane Science, 2016, 509, 125–137. 322 E. Furshpan, The Journal of physiology, 1956, 134, 689–697. 298 H. Yoshida and L. Bocquet, The Journal of chemical physics, 323 M. Kücken, J. Soriano, P. A. Pullarkat, A. Ott and E. M. 2016, 144, 234701. Nicola, Biophysical journal, 2008, 95, 978–985. 299 J. Abraham, K. S. Vasu, C. D. Williams, K. Gopinadhan, Y. Su, 324 J. Küppers, A. Plagemann and U. Thurm, The Journal of C. T. Cherian, J. Dix, E. Prestat, S. J. Haigh, I. V. Grigorieva Membrane Biology, 1986, 91, 107–119. et al., Nature nanotechnology, 2017, 12, 546. 325 J. Fischbarg, J. A. Hernandez, A. A. Rubashkin, P. Iserovich, 300 K. H. Thebo, X. Qian, Q. Zhang, L. Chen, H.-M. Cheng and V. I. Cacace and C. F. Kusnier, The Journal of membrane biol- W. Ren, Nature communications, 2018, 9, 1486. ogy, 2017, 250, 327–333. 301 B. Mi, Science, 2014, 343, 740–742. 326 M. Dvoriashyna, A. J. Foss, E. A. Gaffney, O. E. Jensen and 302 S. Gravelle, H. Yoshida, L. Joly, C. Ybert and L. Bocquet, The R. Repetto, Journal of theoretical biology, 2018, 456, 233– Journal of chemical physics, 2016, 145, 124708. 248. 303 J. Dumais and Y. Forterre, Annual Review of Fluid Mechanics, 327 G. M. Preston, T. P. Carroll, W. B. Guggino and P. Agre, Sci- 2012, 44, 453–478. ence, 1992, 256, 385–387. 304 T. Broyer, The Botanical Review, 1947, 13, 1–58. 328 K. Murata, K. Mitsuoka, T. Hirai, T. Walz, P. Agre, J. B. Hey- 305 M. V. Thompson and N. M. Holbrook, Plant, Cell & Environ- mann, A. Engel and Y. Fujiyoshi, Nature, 2000, 407, 599. ment, 2003, 26, 1561–1577. 329 C. Maurel, L. Verdoucq, D.-T. Luu and V. Santoni, Annu. Rev. 306 K. H. Jensen, J. Lee, T. Bohr, H. Bruus, N. M. Holbrook and Plant Biol., 2008, 59, 595–624. M. A. Zwieniecki, Journal of the Royal Society Interface, 2011, 330 P. Agre, Proceedings of the American Thoracic Society, 2006, 8, 1155–1165. 3, 5–13. 307 H. Rademaker, M. A. Zwieniecki, T. Bohr and K. H. Jensen, 331 H. Javot and C. Maurel, Annals of Botany, 2002, 90, 301– Physical Review E, 2017, 95, 042402. 313. 308 T. Mansﬁeld, A. Hetherington and C. Atkinson, Annual re- 332 M. L. Zeidel, S. V. Ambudkar, B. L. Smith and P. Agre, Bio- view of plant biology, 1990, 41, 55–75. chemistry, 1992, 31, 7436–7440. 309 R. L. Satter and A. W. Galston, Annual Review of Plant Physi- 333 B. Roux, J.-Y. Lapointe and D. G. Bichet, Médecine/Science, ology, 1981, 32, 83–110. 2001, 17, 115–116. 310 W. G. van Doorn and U. van Meeteren, Journal of experimen- 334 A. Horner, F. Zocher, J. Preiner, N. Ollinger, C. Siligan, S. A. tal botany, 2003, 54, 1801–1812. Akimov and P. Pohl, Science advances, 2015, 1, e1400083. 311 R. Hedrich and J. I. Schroeder, Annual review of plant biol- 335 T. Walz, B. L. Smith, M. L. Zeidel, A. Engel and P. Agre, ogy, 1989, 40, 539–569. Journal of Biological Chemistry, 1994, 269, 1583–1586. 312 M. Irving, S. Ritter, A. Tomos and D. Koller, Botanica Acta, 336 T. UIUC, Water channels in cell membranes, YouTube, 1997, 110, 118–126. ”https://www.youtube.com/watch?v=GSi5–y6NHjY”. 313 A. Seminara, T. E. Angelini, J. N. Wilking, H. Vlamakis, 337 E. Tajkhorshid, P. Nollert, M. Ø. Jensen, L. J. Miercke, S. Ebrahim, R. Kolter, D. A. Weitz and M. P. Brenner, Proceed- J. O’connell, R. M. Stroud and K. Schulten, Science, 2002, ings of the National Academy of Sciences, 2012, 109, 1116– 296, 525–530. 1121. 338 H. Sui, B.-G. Han, J. K. Lee, P. Walian and B. K. Jap, Nature, 314 S. Marbach and L. Bocquet, Physical Review X, 2016, 6, 2001, 414, 872. 031008.

42 |Journal Name, [year], [vol.], 1–43 339 S. Gravelle, L. Joly, F. Detcheverry, C. Ybert, C. Cottin- 360 R. K. Soong, G. D. Bachand, H. P. Neves, A. G. Olkhovets, Bizonne and L. Bocquet, Proceedings of the National Academy H. G. Craighead and C. D. Montemagno, Science, 2000, 290, of Sciences, 2013, 201306447. 1555–1558. 340 G. PÃl’rez-Mitta, J. S. Tuninetti, W. Knoll, C. Trautmann, 361 Y. Sowa and R. M. Berry, Quarterly reviews of biophysics, M. E. Toimil-Molares and O. Azzaroni, Journal of the Ameri- 2008, 41, 103–132. can Chemical Society, 2015, 137, 6011–6017. 362 M. Silverman and M. Simon, Nature, 1974, 249, 73. 341 R. Greger and U. Windhorst, From Cellular Mechanisms to 363 S. Kojima and D. F. Blair, Biochemistry, 2004, 43, 26–34. Integration, 1996, 2, year. 364 Y. Nishihara and A. Kitao, Proceedings of the National 342 P. H. Baylis, 1989, 3, 229–578. Academy of Sciences, 2015, 112, 7737–7742. 343 J. L. Stephenson, Kidney international, 1972, 2, 85–94. 365 R. Cereijo-Santaló, Archives of biochemistry and biophysics, 344 A. T. Layton, American Journal of Physiology-Renal Physiol- 1972, 152, 78–82. ogy, 2010, 300, F356–F371. 366 J. H. Miller Jr, K. I. Rajapakshe, H. L. Infante and J. R. Clay- 345 A. Edwards, American Journal of Physiology-Renal Physiol- comb, PLoS One, 2013, 8, e74978. ogy, 2009, 298, F475–F484. 367 N. Y. Yip and M. Elimelech, Environmental science & technol- 346 Gambro, Phoenix X 36 brochure, 2009. ogy, 2012, 46, 5230–5239. 347 J. T. Borenstein, E. J. Weinberg, B. K. Orrick, C. Sundback, 368 I. Statistics, International Energy Agency, 2017. M. R. Kaazempur-Mofrad and J. P. Vacanti, Tissue engineer- 369 R. A. Tufa, S. Pawlowski, J. Veerman, K. Bouzek, ing, 2007, 13, 1837–1844. E. Fontananova, G. di Proﬁo, S. Velizarov, J. G. Crespo, K. Ni- 348 J. C. Kim, F. Garzotto, F. Nalesso, D. Cruz, J. H. Kim, E. Kang, jmeijer and E. Curcio, Applied energy, 2018, 225, 290–331. H. C. Kim and C. Ronco, Expert review of medical devices, 370 S. E. Skilhagen, Desalination and water treatment, 2010, 15, 2011, 8, 567–579. 271–278. 349 P. Armignacco, F. Garzotto, M. Neri, A. Lorenzin and 371 D. MacKay, Sustainable Energy-without the hot air, UIT Cam- C. Ronco, Blood puriﬁcation, 2015, 39, 110–114. bridge, 2008. 350 V. R. Kaila, M. I. Verkhovsky and M. WikstrÃum,˝ Chemical 372 D. J. Rankin and D. M. Huang, Langmuir, 2016, 32, 3420– reviews, 2010, 110, 7062–7081. 3432. 351 P. L. Pedersen and E. Carafoli, Trends in Biochemical Sciences, 373 X. Zhu, J. Hao, B. Bao, Y. Zhou, H. Zhang, J. Pang, Z. Jiang 1987, 12, 146–150. and L. Jiang, Science Advances, 2018, 4, eaau1665. 352 A. Y. Mulkidjanian, K. S. Makarova, M. Y. Galperin and E. V. 374 C.-Y. Lin, C. Combs, Y.-S. Su, L.-H. Yeh and Z. S. Siwy, Jour- Koonin, Nature Reviews Microbiology, 2007, 5, 892. nal of the American Chemical Society, 2019, 141, 3691–3698. 353 M. G. Palmgren, Annual review of plant biology, 2001, 52, 375 D. Brogioli, Physical Review Letters, 2009, 103, 058501. 817–845. 376 D. Brogioli, R. Zhao and P. Biesheuvel, Energy & Environmen- 354 T. Nishi and M. Forgac, Nature reviews Molecular cell biology, tal Science, 2011, 4, 772–777. 2002, 3, 94. 377 M. Simoncelli, N. Ganfoud, A. Sene, M. Haefele, B. Daffos, 355 T. Murata, I. Yamato, Y. Kakinuma, A. G. Leslie and J. E. P.-L. Taberna, M. Salanne, P. Simon and B. Rotenberg, Phys- Walker, Science, 2005, 308, 654–659. ical Review X, 2018, 8, 021024. 356 T. Meier, P. Polzer, K. Diederichs, W. Welte and P. Dimroth, 378 M. Marino, O. Kozynchenko, S. Tennison and D. Brogioli, Science, 2005, 308, 659–662. Journal of Physics: Condensed Matter, 2016, 28, 114004. 357 R. A. Capaldi and R. Aggeler, Trends in biochemical sciences, 379 S. Shin, P. B. Warren and H. A. Stone, Physical Review Ap- 2002, 27, 154–160. plied, 2018, 9, 034012. 358 P. Mitchell, Nature, 1961, 191, 144–148. 380 A. Putnis, Science, 2014, 343, 1441–1442. 359 W. Junge and N. Nelson, Annual Review of Biochemistry, 381 A. Lager, K. J. Webb, I. R. Collins, D. M. Richmond et al., SPE 2015, 84, 631–657. Symposium on Improved Oil Recovery, 2008.

Journal Name, [year], [vol.],1–43 | 43