Einstein and 1905 (2005 was Einstein Year – celebrating 5 key papers published that year.) • In 1905 the atomic hypothesis was not fully accepted. • Despite Brownian motion having been known about for 75 years, its significance was not appreciated. • The kinetic theory of gases was thought of as a 'mechanical analogue', but implied reversibility. • The 2nd law of thermodynamics required irreversibility. • Einstein understood that taking a statistical approach, and assuming atoms existed, reconciled the paradox. • His paper, the second of the 5, was written in April 1905. • • • A macroscopic particle – such as the pollen particle – would be buffeted by the atoms/molecules in the surrounding water. The particle would undergo diffusion and measuring the diffusion constant (or equivalently the displacement) should show an increase with t (not linearly). Perrin's subsequent experiments on sedimentation showed how all this hung together. 34 Brownian Motion Nelson; Dill and Bromberg • • • • Brownian motion of a particle arises due to its constant bombardment by molecules(e.g as in the first observation by Brown in 1827 of pollen grains by the water molecules). Net force, averaged over time, on the particle is zero. But at any moment there is a constantly fluctuating net force and this gives rise to the observed Brownian motion. The motion of the particle follows a random walk • • • • For a sphere of radius r in a liquid of viscosity , we have seen from Stokes Law the force is 6prv This can also be written in terms of the drag coefficient, defined so that Force = drag coefficient × velocity x= 6pr Equation of motion for particle of mass m d 2R dR m 2 x Frandom dt dt where R(t) is the position coordinate, and Frandom is the random force due to collisions. 35 Brownian Motion Jones • Forces are random so each direction behaves in the same way and • We can also use the identity 2 d dx dx d 2x x x 2 dt dt dt dt x 2 y 2 z 2 R 3 x 2 • 2 Work in 1D for simplicity d 2x dx m 2 x Frandom dt dt • Use the substitution d (x2 ) dx 2x dt dt • Then d 2 x x d (x2 ) m 2 Frandom 2 x dt dt • Thus 2 d dx dx x d ( x ) m x m xFrandom 2 dt dt dt dt 2 • Rearranging and taking an average yields x d (x2 ) 2 dt d dx dx xF m x random dt dt dt 2 36 Brownian Motion cont x d (x2 ) 2 dt • d dx dx xF m x random dt dt dt • • • 2 The direction of the random force on the particle is not correlated with the particle's position, so the first term is zero. Similarly, there is no correlation between the particle's position and its velocity, so the second term is also zero. The third term can be rewritten via the theorem for equipartition of energy, since mvx2/2=kBT/2 or vx2=kT/m d x 2 dt And hence the total squared displacement in 3D is R 2 • 3 x 2 6k BT x t The motion is diffusive with a diffusion coefficient given by the Einstein relation D=kBT/x • We will explore the implications of this later, but note the inverse relation between drag (x) and the diffusion constant D. 2k BT x 37 Aside on Ensembles (for the theoretically inclined) Waldram:The theory of thermodynamics • • • • • • On the previous slide we saw averages <….>. What are we averaging over? Those of you who took TP1 will probably assume this is a time average – and it may be. But the ensemble average can also be over a set of replicas at a given instant. In your Thermal Physics course you have been taught about the different types of ensembles. You have also been taught about the Principle of Equal Equilibrium Probability,namely that for an isolated system, all microstates compatible with the given constraints are equally likely to occur. • • • • • If the system obeys this Principle, then an average over the probabilities of the system being in a particular configuration is the same as the average over time. But, formally, one should always specify what sort of average is being taken by the <…..>. Frequently is is simply referred to as an ensemble average, which could mean either. Underlying all this is the ergodic assumption – that the system can go anywhere within the allowed energy range. In which case in equilibrium, the fluctuation distribution is identical with the ensemble average distribution. 38 The Diffusion Coefficient • • When the diffusing particle is a sphere, Stokes Law gives x=6pr, so k BT D 6pr • • • • This is known as the Stokes- Einstein equation. For other shapes the drag coefficient, and hence diffusion constant, take different forms. E.g. disc moving randomly x = 12r Ellipsoid (major and minor axes a and b) x=6pa/ ln(2a/b) for random rotation (other expressions for motion sideways or lengthwise) • • • • For a molecule such as a protein, the radius R may not equate to the actual, physical size. A hydrodynamic radius RH can be defined, which is the effective radius presented to the fluid by the molecule. Additionally, the molecule may or may not permit the fluid to drain through it, depending on the density of chain packing. Thus the detailed hydrodynamics of e.g. globular proteins diffusing through a fluid can be complex. We will return to these concepts when discussing proteins and polymers in more detail. 39 Typical values for D in water Molecule T(oC) MW (g mol-1) D (cm2s-1) Oxygen 25 32 2.1 x 10-5 Sucrose 25 342 5.23 x 10-6 Myosin 20 493,000 1.16 x 10-7 DNA 20 6,000,000 1.3 x 10-8 Tobacco mosaic virus 20 50,000,000 3.0x 10-8 Quantitatively one would expect D ~ m-1/3, but because a hard sphere model for the molecules is not accurate, this precise dependence is not found. 40 Diffusion Equation Dill and Bromberg • Fick's first law states that the flux J of particles is proportional to the concentration gradient. J Dc • • • Specific solutions depend on boundary conditions (recall your 2nd year maths!). e.g for point source diffusing in 1D along x, starting at x=0 And conservation of particles requires c .J t • Hence in 1D and assuming D is a constant, then 2c c D 2 t x • which is Fick's second law in its simplest form. Numbers correspond to values of Dt 41 Diffusion Equation cont • Compare with data from Wall Street • • Distribution of monthly returns for a 100-security portfolio 1945-1970 • • • Note that D need not be a constant, although in many simple situations it is. If it is not, then the equation must be modified to allow for the fact that D can be a function of position. c c D( x) t dx x Over time stock prices also exhibit a random walk with drift! Individual whims can lead to the statistical movement. Baseline drift comes from the fact that overall shares do make a profit…. 42 Diffusion Control • Diffusion may limit: 1. – Growth of 2nd phase particles – Supply of nutrients to organisms – Colloidal aggregation a • • • We will model as spherically symmetric, so work in spherical coordinates. Assume steady state so dc/dt=0. Diffusion equation becomes 1 d 2 (rc) c 0 2 r dr 2 • Solution depends on boundary conditions. Diffusion of molecules which react at the surface. If these are transformed/lost during the reaction, then c(r = a) = 0. If the concentration well away from particle is c solution is a c(r ) c (1 ) r • For such a concentration profile, the flux is given by dc Dc a J (r ) D 2 dr r 43 Diffusion Controlled Collisions and Reactions • Hence the number of collisions at the surface per unit time I(a) is 2 • I (a) J (a)4pa 2 4pDc a • • • • • Minus sign implies flux is towards particle (-r direction). This is the fastest possible rate at which any process can occur, i.e. when it is limited by diffusion. If the reaction at the surface itself is rate limiting, a process will be slower. If the diffusing molecule is a nutrient to an organism, then we also have to think about its consumption and this will be proportional to the volume. Thus if the bacterium etc is too large, cannot get sufficient supply of the nutrient by diffusion, and so this will set an upper limit on size. • Growth of particles post-nucleation In this case the molecules diffusing to the surface are causing growth of the particle – a process known as Ostwald ripening or coarsening. Flux J(r) as before but now we have J (r ) • 1 dr r2 dt Integration yields r3 t or r t 1/ 3 • This is known as the Lifshitz-Slyozov law, and is empirically found to hold as long as diffusion is the limiting factor. 44 Diffusion Controlled Aggregation • No salt • • With salt • • • • During colloidal aggregation, can have either diffusion or 'reaction' being the limiting factor. Which is dominant depends on the conditions of aggregation E.g. in the examples shown here of 250nm polymethyl methylacrylate particles, it is the salt concentration which is changing. This changes the interparticle potential. The shape of the aggregates can provide insight into the processes occurring. Very often fractal structures form. In these the structure is the same at all lengthscales R~M 1/ d f Watching paint dry; a new paint formulation; • Where R is the size, M is the mass and the fractal structures fractal dimension 1<df3 (equality for compact again visible, but object) clearly with 45 rearrangement Aside on Fractals Jones • • • • Computer simulations show such fractals should have a fractal dimension of 1.71 when diffusion limited aggregation occurs. • If clusters are allowed themselves to diffuse, the dimension is larger at 1.78 implying a slightly more dense structure. • In practice, the rate at which rearrangements and sticking occur will all affect the fractal dimension. • But it always lies between 1 and 3. • When aggregation occurs – of individual molecules or particles – the structures that result depend on the probability of sticking i.e. whether diffusion or sticking is the limiting factor. If rearrangements can then occur the structure will be able to compact. If not, a fractal structure results, over a range of lengthscales from the size of the particle up to the size of the aggregate. 46 Implications of the Einstein Relation Dill and Bromsgrove • We have derived the Einstein equation D • • • k BT In this, the drag coefficient x (dissipation) and the diffusion coefficient D (fluctuations about equilibrium) are inversely related. We will see shortly that this is a very general result. In general, if there are random fluctuating forces f(t) and a drag force xv acting together on a particle, we can write The ensemble average of v is given by m x dv m f (t ) xv dt • • • • dv f (t ) x v(t ) dt <f(t)> =0 since this is a fluctuating force arising from the many collisions. So the solution for <v(t)> is v(t ) v(0) exp x t / m • So we can define a characteristic time t = m/x a viscous relaxation time. • Example: For a sphere of radius 10nm (typical size for a large polymer) in water, this leads to t ~10-10 s. This is known as the Langevin equation (strictly the A-Langevin equation, as dealing with acceleration). 47 Velocity Correlation Function (Doi, Balescu – see the latter for a rigorous and general discussion) • • • • • • More generally, look at the velocity correlation function defined as <v(0)v(t)>. Recall the <…> imply a suitable ensemble average. Correlation functions in general measure the extent to which there is any correlation between a function at two separate times. It can be applied to position, force, velocity etc, and can be defined in the same way as for the velocity correlation function. The larger the value of the function is, the greater the extent of correlation. For random motion, the velocity correlation function equals <v2(0)> for t~0, and = 0 at large t, when all correlations lost. • • For so-called stationary stochastic processes, any autocorrelation function only depends on the time interval involved i.e <v(t1)v(t2)> = Cv(t1- t2) with Cv a function Thus more specifically we can write v(t )v(0) 2D (t ) • • • Where D is the diffusion coefficient. This can be seen by considering, the mean square displacement (MSD) in 1 dimension about an average position x = 0 for diffusive motion is t t 0 0 t t 0 0 x(t ) 2 dt1 dt2 v(t1 )v(t2 ) x(t ) dt1 dt2 2 D (t1 t2 ) 2 Dt 2 48 Velocity Correlation Function cont • Returning to the Langevin equation m • dv f (t ) xv dt By equipartition of energy 1 1 m v 2 ( 0) k B T 2 2 kT v(0)v(t ) B exp (x t / m) m multiply by v(0) and take the ensemble average Velocity correlation function • dv(t ) v(0) f (t ) x v(0)v(t ) dt Now v(0) and f(t) are uncorrelated so that the first term on RHS = 0, and d x v(0)v(t ) v(0)v(t ) 0 dt m v(0)v(t ) v 2 (0) exp (x t / m) • <v2(0)> <v(0)v(t)> m v(0) So we have obtained an equation for the velocity autocorrelation function. • • <v2(0)>e-xt/m t The figure confirms that the longer the time interval, the less the correlation between v(0) and v(t) . For timescales shorter than t the characteristic time) the velocity retains correlation, but not for time scales much longer than this. 49 Fluctuation-Dissipation Theorem Dill and Bromsgrove • • Conceptually, the fluctuation-dissipation theorem states that the fluctuations in a system are correlated (inversely) with the energy dissipation, so generalises our conclusion from the Einstein equation. In the Einstein relation D • • • k BT x we see that the drag coefficient x (dissipation) and the diffusion coefficient D (directly related to the fluctuations in mean position as we have just seen) are inversely related. Large dissipation leads to small fluctuations about equilibrium. • • We have seen that the velocity autocorrelation function indicates how fast a particle 'forgets' its initial velocity due to the effect of Brownian motion and collisions leading to randomisation. If this timescale is long, then clearly there is little dissipation, there are few collisions, and equilibrium is slow to be achieved. Thus low dissipation means it takes a long time to establish equilibrium. t is m/ x and is typically in the picosecond range for a small protein. 50 Fluctuation-Dissipation Theorem cont 0 • • • • Integrate the time correlation function over all possible lag times t <v2(0)> k BT k BT v(0)v(t ) dt exp ( x t / m ) dt D m 0 x This integral equals the area under the curve. If the area is small it implies that kT/x is small and D is also small: the dissipation is large, and the diffusion coefficient/ transport coefficient is small. Equilibrium is rapidly reached as there are many collisions. And the fluctuations about equilibrium are small, as we saw from the MSD. <v(0)v(t)> • t • • • Conversely, if the area is large, the dissipation is small, and the fluctuations are large. So we have a statement about an equilibrium property – the fluctuations – related to a non-equilibrum property , in the form of dissipation. The magnitude of equilibrium fluctuations is related to how fast the system reaches equilibrium. 51 Further thoughts on the Fluctuation Dissipation Theorem cont • • • • • Further Examples of how the FDT applies Consider a pendulum moving about an equilibrium position, due to fluctuations in air compared with its motion in a viscous fluid. At a fixed temperature, the mean square displacement will be the same in both cases, corresponding to the magnitude of the fluctuation in the displacement. However since the damping in the second case is much greater, then the random forces f(t) giving rise to the fluctuations must be greater too. An alternative form of the equation expressing this relationship is given by 1 x 2k BT • Nyquist's formula in electrical circuits says that the larger the resistance (giving rise to dissipation) the larger the (thermal) noise emf present. • Thus this theorem translates into many different situations where fluctuations and dissipation are present. • BUT It only applies to equilibrium systems. • Recent work has been directed at trying to understand how far it can be pushed e.g. for glasses which have quenched in disorder f (0) f (t ) dt force correlation function 52 Using Brownian Motion to obtain Local Viscosities • • • • Beads moving around in viscous fluids provide a means of probing local viscosities. In some materials (e.g. liquid crystals, see later) these may be anisotropic. Other systems (e.g gels) may be heterogeneous. A generalised version of the Stokes Einstein equation allows the complex moduli to be determined by tracking the motion of the particles. 53 Using Particle Tracking in Practice • mm The mean square displacement as a function of time can be followed eg. during some process. dk BT r (t ) t 3pa 2 • mm mm • • mm (d is dimensionality, a is particle radius, t is lag time) Data for hectorite clay 1 hour and 24 hours after the start of the experiment, showing how Brownian motion becomes restricted as the systems 'sets' or 'gels'. From the displacements the viscoelastic moduli can be determined (via Laplace transforms). dk BT G (s) 3pas r ( s ) 2 54 Chemical Potential and Free Energy • Recall the definition of the Helmholtz free energy F for a simple system • F=U-TS • • If one has a reservoir containing N atoms/molecules, the chemical potential m is defined as F m N T ,V Model situation: A • • B Particle (species i) and energy exchange is possible between two boxes A and B. Total change in entropy The chemical potential is the change in free energy when one particle moves from one position, phase etc to another. Equivalently S S dS tot A dE A B dE B E A E B G U H S m T N N N N T , P S ,V S ,P E ,V S A S B i N dN Ai i N dN Bi Ai Bi • where summation is over species i. • In this field of soft condensed matter, F and G are often used interchangeably since the PV term is usually negligible. 55 Chemical Potential and Equilibrium • • Now both energy and particle number must be conserved, and the net change in entropy must be zero or greater. Using the relationship between m and S and recall 1/T = dS/dE and dEA+dEB=0 this equation reduces to 1 1 m m dE A Ai Bi dN Ai 0 TA TB i TA TB • and this must be true for all the species. This implies (obviously) that at equilibrium TA=TB and • • When systems can exchange particles, not only the temperature but also the chemical potential for each species must be equal at equilibrium. Can now extend all the thermodynamic functions to include the possibility of Ni varying. dF SdT pdV mi dN i i dH TdS Vdp mi dN i i dG SdT VdP mi dN i i m A= m B 56 Aside on the Clausius Clapeyron Equation • Liquid P Since we must have mgas=mliquid on the phase boundary, then since dG = 0 also across the boundary Solid Vapour T • Recall this tells us about how properties change along the phase boundary, eg between gas and liquid. S gasdT Vgasdp SliqdT Vliqdp • But across the boundary T and p must be equal so dp S L dT V TV 57 Chemical Potential of a Perfect Gas • • Recall F= -kBTlnZ where Z is the partition function. • z zk BT N pV For N indistinguishable particles, as in a perfect gas, Z = zN/N! • Thus m = mo + kBTlnP where mo represents the chemical potential of a standard state. • If you have a solution in equilibrium with its vapour msoln= mvap • Hence • o Equivalently m so ln m k BT ln c • Where c is the concentration in the solution. This result is generally true. where z is the individual particle partition function. • Now z/N can be written in terms of the pressure as Then z F k BT N ln N N Hence F z k BT ln N T ,V N m • m so ln m o k BT ln P 58 Membranes • • • • • Membranes are the envelopes around all our cells in our bodies. They are complex, dynamic structures comprising so-called lipid bilayers (see later). They have to be able to respond to significant shape changes as our bodies move or an organism crawls. Initially let us treat the membrane as a simple fluid interface with associated surface tension, and look at its fluctuations. This is far too simple, because in fact there is a curvature elasticity to be taken into account (to be discussed later). Vesicles – simple approximation to biological cells - easily distort to allow for shape changes, but do not exhibit short wavelength fluctuations as these cost a lot of energy. 59 Monge Representation of a Membrane Surface Boal • • • The Monge representation is a common way of representing an undulating surface, such as a membrane. A point r on a surface is represented by r = (x,y, h(x,y)) • We can construct tangent vectors along x and y, represented by rx and ry by taking a unit step in the x (or y) direction, and a step h hx x h or hy y in the z direction, to give rx (1, 0, hx ) ry (0, 1, hy ) z y x • • h(x,y) h represents the 'height' away from the x,y surface. For complex surfaces, h may not be single valued (e.g if overhangs), but in the Monge representation it is constrained to be single-valued. • • These are not unit vectors, and are not generally orthogonal. They define the plane tangent to the surface at (x,y,h(x,y)) by r r h ,h ,1 n x y rx ry 1 h x y 2 x 2 1/ 2 y h 60 Monge Representation cont • n rydy z hx x • rxdx 1 • 2r C n. 2 s The element of area dA is defined by dA rx ry dxdy 1 h h • dA 2 x 2 1/ 2 y And the metric g of the surface is defined by g (1 hx2 hy2 ) The surface can also be defined by its curvature – these are simply geometrical relationships, discussed in maths texts. The curvature of a surface is defined as dxdy • • • where s is the arc length. In general there will be two principal curvatures (the two extremal values) C1 and C2. The combination (C1+ C2) is the mean curvature . And C1. C2 is the Gaussian curvature . So that dA=g dx dy 61 Fluctuations in Membranes treated as a Simple Fluid Safran • h(r ) h(q)e iq.r h(x,y) q h(q) h(r )e iq.r dr fluctuating surface • • • For a simple fluid surface there will be an associated surface energy g per unit area. The change in free energy associated with the undulations of the surface h(x,y) is g where 2 2 F (h x hy )dxdy 2 h h hx ; hy x y Hence, writing the undulations as a Fourier sum • The free energy F is given by F g q 2 2 | h(q) |2 q • And therefore by equipartition of energy (for each mode q) k BT | h(q) | 2 gq 2 62 Interpreting these Fluctuations Safran • Consider h 2 (r ) • • • • • 1 (2p ) 2 | h(q) |2 dq • Thus the mean square fluctuation diverges as the logarithm of the system size. Also, since k BT | h(q) | 2 gq 2 In two dimensions the logarithm of the integral diverges at small q, and so we must think carefully about the limits. Set the upper limit as p/a, where a is the atomic size. Set the lower limit as p /L, where L is the size of the interface. Then k T L h (r ) B log 2pg a 2 • • small wavelength fluctuations in real space (ie large q) have far smaller fluctuations associated with them than large wavelength (small q) fluctuations. Or equivalently higher energy is associated with small wavelength fluctuations. Thus largescale distortions of membranes are more favourable than local rumpling. 63 Real Cells • • • • • Bone cells • • • Example: Bovine pulmonary arterial endothelial cells imaged in the confocal laser scanning microscope Dual labeled with a green stain for actin microfilaments (FITC-Phalloidoin) and a red stain for mitochondria. The stain for actin is actually derived from a toxic compound found in mushrooms. The stain for mitochondria only becomes fluorescent when the stain is activated by enzymes which reside in the mitochondria. For this reason, only the mitochondria appear red, not the other organelles. Example: Bone cells imaged in the environmental scanning electron microscope (ESEM). Normal scanning electron microscopes work in high vacuum, and so require cells to be dehydrated. The ESEM does not, and so potentially provides a new route to high resolution examination of cells. 64 Real Cells cont • • Real cells are a lot more complex than simply an interface undergoing fluctuations, but the basic ideas are correct. Have to take into account – The physical structure of the membrane – The (bio)chemical structure of the membrane – The contents of the cell, particularly the cytoskeleton. – What is surrounding the cell – Osmotic effects Example: what happens when the cell is placed in different media which cause different degrees of stress? The example shown is for changing the osmolarity of the surrounding medium, when the cell shrinks or stretches according to whether molecules have flowed in or out. 65 Aside on Osmotic Pressure • • • The comment on osmolarity introduces the idea of osmotic pressure. You will be familiar with this through the idea of what happens with a semi-permeable membrane. A • B • pV RT ci • Which for low concentrations approximates to p • • • If there is solute in A, then water will flow across the semi-permeable membrane until the chemical • potentials on the two sides are equal. m mixture (T , p p , ci ) m pure (T , p,0) The osmotic pressure, p, is the pressure which will just stop the flow occurring. In general N RT V This has a form equivalent to the case of a perfect gas. This effect is crucial in maintaining cells healthy. For instance, when a plant 'wilts' it has lost water from its cells and the external pressure causes the cells to collapse. By watering the plant, water is sucked back into the cells to return the plant to full 'turgor' (in this context one talks about 'turgor pressure', and the cells are then said to be turgid). 66 Brownian Motion within Cells- current (and preliminary!) Research Two dimensional mean square displacements of beads in this cell. A selection of particles have been highlighted in red, for clarity. The slopes exhibit significant scatter, Fluorescence image of an individual which we attribute to local human dermal fibroblast cell showing variations in the rheology according67 clustering of the beads in the perinuclear region, with a few beads in the periphery of to cytoskeletal arrangements. (Preliminary) Microrheology It is clear very different responses can be measured. From the mean square displacements one can then extract the local moduli, and examine how anisotropic they are. The aim is now to relate these local measurements to the internal components of the cell under different conditions e.g of cell adhesion, motility, stress etc. We do not yet know if the variation is real, and we will be able to make these correlations, or simply due to a spread in values for statistical 68 reasons.