Discover · No algebra required

A particle wanders.
What should you measure?

If heat really is the motion of molecules, then something small enough to see under a microscope has to be shoved about hard enough to watch. Einstein turned that thought into a number a laboratory could go and check. You can arrive at the same number here, deciding at each step what is worth measuring.

A route you could take, not a transcript of Einstein’s private thoughts. Every step can be read without running a simulation.

The 1904 shelf

What a careful reader of the literature had by the end of 1904, and nothing later unless it is marked as parallel work. The route uses nothing else.

This list is a reconstruction aid assembled for learning, not a documentary history of what Einstein read.

Legend:Available by the end of 1904Parallel work: not available to a 1904 readerAdmitted 1905 import
Published, 1828-09

Particles contained in the pollen of plants, and still smaller particles from every mineral Brown could powder finely enough to stay suspended in water, are seen in motion in the water; he found the motion in plants dried for a century as well as in living ones.

Available by the end of 1904
Limits: Brown ruled out currents in the fluid and its evaporation, said the motion belonged to the particle itself, and declined to hazard any conjecture about the molecules. What he reports persisting is the motion in dead and century-old material; he does not say how long one particle keeps moving, and does not call the motion irregular.

Historical context

Available in the published scientific literature or public proceedings prior to 1905.

Available by

Published, 1828-09

Prior event: Performed 1827-08

What the paper itself cites or asserts

Not yet recorded for this card.

Where this site uses it

Steps 1 and 2 of this route

Primary evidence and sources

  • A brief Account of Microscopical Observations made in the Months of June, July, and August, 1827, on the Particles contained in the Pollen of Plants; and on the general Existence of active Molecules in Organic and Inorganic Bodies Phil. Mag. (new series) 4 (1828) 161 (1828-09)
Published 1851

A sphere of radius a moving slowly and uniformly at speed v through a fluid of viscosity η is held back by a force F = 6πηav.

Available by the end of 1904
Limits: The letters are later ones: Stokes prints the resistance as 6πμ′ρaV, where μ′, his index of friction, is the viscosity divided by the density ρ. He gives it only for motion so slow that the square of the velocity can be neglected, for a fluid that does not slide past the surface of the solid, and says the fluid may be liquid or gas.

Historical context

Available in the published scientific literature or public proceedings prior to 1905.

Available by

Published 1851

Prior event: Presented 1850-12-09

What the paper itself cites or asserts

Not yet recorded for this card.

Where this site uses it

Step 3 of this route

Primary evidence and sources

  • On the Effect of the Internal Friction of Fluids on the Motion of Pendulums Trans. Camb. Phil. Soc. 9, part II, p. [8]; the sphere, eq. (126), p. [51] (1851)
Published 1855

Dissolved matter moves down its concentration gradient at a rate proportional to the gradient; with conservation of matter this gives a diffusion equation for the concentration.

Available by the end of 1904
Limits: Fick could not derive the law from the general laws of motion. He proposes it by analogy with Fourier's law for the flow of heat, as a conjecture he says his experiments put beyond doubt, and writes it for a salt solution in a vessel, with a constant that depends on the substances.

Historical context

Available in the published scientific literature or public proceedings prior to 1905.

Available by

Published 1855

What the paper itself cites or asserts

Not yet recorded for this card.

Where this site uses it

Step 3 of this route

Primary evidence and sources

  • Ueber Diffusion A. Fick, Ann. Phys. (Pogg.) 94, 59 (1855)
Published, 1860-01

For a model gas of perfectly elastic spheres, particles of different masses moving in one vessel come after many impacts to the same mean vis viva, mass times mean square velocity. Explaining the pressure of a gas by assuming the square of the velocity proportional to the absolute temperature, Maxwell concludes that equal volumes of gases at one pressure and temperature hold equal numbers of particles.

Available by the end of 1904
Limits: Shown for hard, perfectly elastic spheres, which Maxwell offers as a physical analogy to be tested against real gases. The link to the absolute temperature is an assumption that accounts for the pressure, not something the paper derives.

Historical context

Available in the published scientific literature or public proceedings prior to 1905.

Available by

Published, 1860-01

Prior event: Presented 1859-09-21

What the paper itself cites or asserts

Not yet recorded for this card.

Where this site uses it

Steps 1 and 2 of this route

Primary evidence and sources

  • Illustrations of the Dynamical Theory of Gases. Part I. On the Motions and Collisions of Perfectly Elastic Spheres Phil. Mag. (4) 19, 19 (1860-01)
Published 1879

Nägeli estimates the speed one molecular impact can give a starch grain 3 micrometres across, finds it far too small to see even under the microscope, reckons that about a million molecules would have to strike at once in one direction to make a single visible jerk, and concludes that the motions of the liquid's molecules cannot be what makes the particles dance.

Available by the end of 1904
Limits: He could not calculate the effect of a water molecule, whose speed was unknown, so he used a water-vapour molecule in air as an upper bound. He did consider the very many impacts, more than a trillion a second, and held that, coming from every side, they cancel completely; he does not estimate how large their chance imbalance over a short interval would be, which is where the route turns. For the cause he points instead to attracting and repelling forces between molecules, perhaps electric, and calls that only a possibility.

Historical context

Available in the published scientific literature or public proceedings prior to 1905.

Available by

Published 1879

Prior event: Presented 1879-06-07

What the paper itself cites or asserts

Not yet recorded for this card.

Where this site uses it

Step 2 of this route

Primary evidence and sources

  • Ueber die Bewegungen kleinster Körperchen C. v. Nägeli, Sitzungsber. math.-phys. Cl. Akad. Wiss. München 9 (1879), p. 389 (1879)
Published 1887

A substance dissolved in a solution presses on a wall that lets only the solvent through, and that osmotic pressure obeys the gas laws of Boyle, Gay-Lussac and Avogadro: van 't Hoff writes PV = RT, with one value of R for every gas and every dissolved substance when each is counted in molecular weights.

Available by the end of 1904
Limits: Only when the solution is dilute enough that the dissolved particles' interaction and volume can be neglected, the restriction that holds for gases too. In water most salts and the strong acids and bases give a larger pressure, i times the expected value, which he ascribes, on Arrhenius's suggestion, to splitting into ions. Whether the law holds for particles large enough to see is the question the route asks, not something the law says.

Historical context

Available in the published scientific literature or public proceedings prior to 1905.

Available by

Published 1887

What the paper itself cites or asserts

Not yet recorded for this card.

Where this site uses it

Steps 2 and 3 of this route

Primary evidence and sources

  • Die Rolle des osmotischen Druckes in der Analogie zwischen Lösungen und Gasen J. H. van 't Hoff, Z. physik. Chem. 1 (1887) 481 (1887)
Published 1888

The motion is intrinsic and persistent: no accidental effect of currents, vibrations or differences of temperature, it goes on at constant temperature and indefinitely while the particles stay suspended. It is livelier the smaller the particles, increases with temperature, and varies with the liquid.

Available by the end of 1904
Limits: Gouy does not speak of viscosity; of the liquids, he says pure water seems to show the motion most strongly. He reads it as showing not the molecules' own motions but something closely tied to them, and as direct, visible evidence for the kinetic hypotheses about heat. In a note he doubts that uncoordinated molecular motions produce it directly, since they should move only far smaller particles.

Historical context

Available in the published scientific literature or public proceedings prior to 1905.

Available by

Published 1888

What the paper itself cites or asserts

Not yet recorded for this card.

Where this site uses it

Step 1 of this route

Primary evidence and sources

  • Note sur le mouvement brownien Gouy, J. Phys. Théor. Appl. (2) 7, 561 (1888)
Published, 1900-08-09

Exner traced the paths of gamboge particles in water for half a minute to a minute and measured their apparent speeds: lower for larger particles, higher at higher temperature. Giving the particles the same kinetic energy as the liquid's molecules, those speeds imply a molecular speed of about 30 cm a second at 20 °C, nowhere near the 270 metres a second that G. Jäger had calculated for 0 °C.

Available by the end of 1904
Limits: He says the values are not very accurate and probably too small, because the smallest excursions could not be traced, and that the simple assumption of particles behaving like the liquid's molecules does not suffice. He still leaves open that these visible motions may one day give measures of the liquid's inner motion.

Historical context

Available in the published scientific literature or public proceedings prior to 1905.

Available by

Published, 1900-08-09

What the paper itself cites or asserts

Not yet recorded for this card.

Where this site uses it

Step 1 of this route

Primary evidence and sources

  • Notiz zu Brown's Molecularbewegung F. M. Exner, Ann. Phys. (4) 2, 843 (1900-08-09)
Published, 1902-12-30

Siedentopf and Zsigmondy light small particles with a beam at right angles to the microscope's line of sight, so that no illuminating ray enters the objective; particles far smaller than half a wavelength of light, below what a microscope can resolve, then show one by one as bright diffraction discs on a dark field.

Available by the end of 1904
Limits: Their subject is gold particles in ruby glass; they report having begun on colloidal solutions and turbid media. A particle is seen, not imaged: its size is estimated separately, not read off the picture.

Historical context

Available in the published scientific literature or public proceedings prior to 1905.

Available by

Published, 1902-12-30

What the paper itself cites or asserts

Not yet recorded for this card.

Where this site uses it

Step 1 of this route

Primary evidence and sources

  • Über Sichtbarmachung und Größenbestimmung ultramikroskopischer Teilchen, mit besonderer Anwendung auf Goldrubingläser H. Siedentopf and R. Zsigmondy, Ann. Phys. (4) 10, 1 (1902-12-30)
Presented, 1904-01-12

William Sutherland's paper on the measurement of large molecular masses, communicated at the Dunedin meeting of the Australasian Association for the Advancement of Science in January 1904, uses Stokes's formula for the resistance on a sphere to make a dissolved molecule's coefficient of diffusion inversely proportional to its radius and to the viscosity of the liquid.

Available by the end of 1904
Limits: No correction for slip: that first appears in his June 1905 paper in the Philosophical Magazine. The paper itself finds the diffusion data better represented by the inverse square root of the molecular volume than by the cube root its formula gives. Its text is known from the Association's Report, printed in Wellington in 1905; the sources say the paper was communicated, not who read it.

Historical context

Available in the published scientific literature or public proceedings prior to 1905.

Primary evidence and sources

  • The Measurement of Large Molecular Masses Report of the Tenth Meeting of the Australasian Association for the Advancement of Science, Dunedin, 1904 (Wellington, 1905), p. 117 (1905)
Published, 1905-06

Sutherland publishes in the Philosophical Magazine a formula for the coefficient of diffusion of a molecule treated as a sphere, now corrected for slip at its surface through a coefficient of sliding friction; with no slip it reduces to the gas constant times the temperature divided by six pi times the viscosity, the radius and the number of molecules in a gram-molecule.

Parallel work: not available to a 1904 reader
Limits: The slip correction is new here: the paper opens by recalling that the Dunedin formula of 1904 made diffusion vary inversely as the radius and the viscosity, and adds the correction after looking more closely at the dynamics.

Historical context

This work appeared alongside or after Einstein’s 1905 paper and was not available to a 1904 reader. The June 1905 Philosophical Magazine publication falls between Annalen's receipt of Einstein's paper on 11 May 1905 and its publication on 18 July 1905.

Primary evidence and sources

  • A Dynamical Theory of Diffusion for Non-Electrolytes and the Molecular Mass of Albumin Phil. Mag. (6) 9, 781; the formula, eq. (3), p. 782 (1905-06)

The nagging fact

If Gouy is right and the jiggling is thermal, then a particle a micron across is only a very large molecule, and what is known about molecules in solution should hold for it too. Does it?

The first honest question

Would a particle you can see, suspended in water, press on a membrane the way a dissolved molecule does?

01 / Choose a quantity

Averages can hide motion.

Imagine marking each particle’s starting point and looking again after one second. A movement to the right is positive; a movement to the left is negative. When left and right are equally likely, these signed displacements cancel on average, even though the particles have spread out.

Before you read on, choose what you would measure: the signed displacement, its size regardless of sign, or its square. The last two both keep the information about the spread.

Why use the squared displacement?

Squaring makes leftward and rightward displacements alike positive. It also has a property the absolute value lacks: for independent steps with zero mean, the mean squares add. That is why the square, and not the absolute value, carries the derivation.

02 / Ask whether it presses

Does a particle you can see push like a molecule?

In 1887 van ’t Hoff showed that a substance dissolved in a dilute solution presses on a membrane that lets only the water through, and that this osmotic pressure follows the gas law. The law counts molecules. Nothing in it says how big a molecule is.

So take the nagging fact at its word. If a particle a micron across is only a very large molecule, the same law should hold for it, with the number of particles in place of the number of molecules. The usual view in 1904 said otherwise: a membrane holding back suspended particles should feel no force at all, because their free energy did not seem to depend on where they are. The kinetic theory says it must feel one. Both cannot be right, and the particles’ wandering is where they part.

Why the particle’s size drops out

The paper’s §2 finds the free energy of the suspension without solving any molecule’s motion. All it needs is how the particles share the volume that holds them, and that brings in how many there are, not how big they are. The pressure comes out as the gas law for that many particles:

p=RTN νp=\frac{RT}{N}\,\nu

Here ν is the number of particles per unit volume, and N is the number of molecules in a gram-molecule: the number this route ends by estimating.

Two ways on

Can molecular impacts move a particle large enough to see?

The branches vary which theoretical principle is taken as a starting postulate.

No single impact is big enough

Held by Carl Nägeli, 1879

Constrained by physical contradiction

Hypothesis
A water molecule is so much lighter than a particle a micron across that one impact changes the particle's speed by far too little to see. Impacts arrive from every side, so they cancel, and the molecules cannot be what moves the particle.
Where it holds
For one impact on its own the estimate is right: the particle's speed barely changes.
Its steps (3)
  1. Estimate what one impact does. The molecule's momentum, shared with a particle many billions of times heavier, changes the particle's speed by an amount no microscope could follow.
  2. Conclude that the impacts, arriving from every side, cancel, so that in still water a particle stays where it is.
  3. Follow the consequence. Particles that nothing moves cannot spread, so a step in their concentration stays a step for ever. The laboratory linked below runs that world with the kicks turned off.

The estimate for one impact is sound, and the conclusion does not follow from it. With the kicks off nothing spreads, which contradicts Fick's law for dissolved matter and Gouy's observation that the motion never dies away.

The imbalance of many impacts moves it

The route taken in the 1905 paper

Hypothesis
In any short interval a particle is struck an enormous number of times. The impacts do not cancel exactly, and the imbalance left over, different in each interval, moves it by an amount you can see over a time you can wait.
Where it holds
When the particle is watched over intervals long compared with the time between impacts, and the imbalances in successive intervals are independent of one another.
Its steps (2)
  1. Ask how far the particle gets over an interval, not what one impact does.
  2. Treat its displacements in successive intervals as independent, each as likely to go one way as the other. Their squares add, so the mean square grows in proportion to the time.

This is the paper's route in §4, and it is the move marked below: the products of different displacements average away, so the mean squares of independent displacements add.

03 / Balance a force against drag

Pull on the particles, and see what holds them back.

Suppose a force pulls every particle the same way, toward the bottom say. In equilibrium the particles crowd together until their osmotic pressure pushes back as hard as the force pulls. That is one account of the crowd.

Here is another. Under the force each particle drifts through the water, held back by Stokes’s drag, while the crowding makes the particles diffuse back from more to fewer, as Fick’s law says. In equilibrium the drift one way and the diffusion the other carry equal numbers across each second.

Put the two accounts side by side and the force cancels: a stronger force makes a steeper crowd in exactly the proportion that it makes a faster drift. What is left fixes the diffusion coefficient by the temperature, the viscosity of the liquid and the radius of the particle alone.

The coefficient, in the paper’s letters and in today’s
D=RTN 16πkPD=\frac{RT}{N}\,\frac{1}{6\pi kP}

Einstein writes k for the viscosity and P for the radius. In today’s letters the same relation reads as below, with Boltzmann’s constant kB = R/N. The paper’s k is not Boltzmann’s constant.

D=kBT6πηaD=\frac{k_BT}{6\pi\eta a}

This is a side door into the paper: it reaches D without following any one particle. The next steps follow one particle, and they reach the same D.

04 / Make a prediction

Four times as long. Four times as far?

Wait four times as long. Does a particle get no farther, twice as far, or four times as far? It depends on how it moves. A particle drifting at a steady speed would go four times as far. A particle kicked about at random, each kick independent of the last, would not. Choose before you open the argument.

Why a randomly kicked particle goes only twice as far

Four times the observation time gives four times the mean square displacement. Taking its square root gives twice the RMS displacement. The square grows linearly with time; its square root does not.

⟨x2⟩=2Dtλx=2Dt\langle x^2\rangle=2Dt \qquad \lambda_x=\sqrt{2Dt}

The laboratory’s one-second, ten-second and one-minute comparison is calculated from one diffusion coefficient. It is not three independent measurements.

Put it in your own words. In four times as long, a particle kicked about at random typically gets only twice as far. Why not four times as far, as a particle drifting at a steady speed would?

What you write stays in this box. Nothing on the page reads it, marks it or sends it anywhere, and it is gone when you leave.

Compare with a worked explanation

Only a person can judge an explanation, and here that person is you. Read yours against each of these in turn:

  • Each kick is independent of the last, so a step one way is as likely to be undone by the next step as to be continued.
  • What grows in proportion to time is the mean square displacement, because the mean squares of independent steps add.
  • The typical distance is the square root of the mean square, so four times the time gives the square root of four: twice as far.
  • A drifting particle’s steps all point the same way, so nothing cancels and its distance grows in proportion to time.

Split the time into many short intervals. In each one the particle is kicked one way or the other, independently of every earlier kick, and its displacement is the sum of these steps. The average of that sum is zero, because each direction is equally likely, but the average of its square is not. Squaring the sum gives each step’s own square plus products of pairs of different steps, and because the steps are independent each of those products averages to zero. What is left is the sum of the steps’ own squares, which grows with the number of steps, and so in proportion to time. Four times as long gives four times the mean square, and the typical distance, its square root, is the square root of four times as large: twice. A particle drifting at a steady speed takes steps that all point the same way. Nothing cancels, and four times the time carries it four times as far.

Two ways on

What should you measure: how fast a particle moves, or how far it gets?

The branches vary what physical quantity is chosen for measurement.

Measure its speed

Held by Felix Exner, 1900

Constrained by physical contradiction

Hypothesis
Trace the particle's path under the microscope for half a minute to a minute and divide the distance by the time. If molecules shove the particles, their energy of motion should match the molecules' at the same temperature, and the speed tells you whether it does.
Where it holds
For a particle drifting at a steady speed, where every interval gives the same answer. And as a stated quantity: a speed over a named interval is something you can measure and report.
Its steps (3)
  1. Measure the displacement over an interval τ and divide by τ. Exner did this with care, each value a mean of about ten tracings of half a minute to a minute. Given the kinetic energy of the liquid's molecules, his particles implied molecules moving about 30 cm a second at 20 °C, where G. Jäger had calculated 270 metres a second.
  2. Now change τ. For a particle kicked at random the displacement grows only as the square root of τ, so the displacement divided by τ grows as the interval shrinks: a quarter of the interval, twice the speed.
  3. So the number depends on how often you look, and it has no limit as you look more often. It is not a property of the particle, and comparing it with the speeds of molecules compares nothing.

The measurements can be right and the comparison still fails: for a randomly kicked particle, an apparent speed is set by the interval as much as by the particle.

Measure how far it gets in a given time

The route taken in the 1905 paper

Hypothesis
Mark where the particle starts, look again after a fixed interval, and record the displacement. Repeat, and take the mean square.
Where it holds
Whenever the displacements in successive intervals are independent. Then the mean square grows in proportion to the interval, and its ratio to the interval belongs to the particle and the liquid, not to the observer.
Its steps (2)
  1. Fix the interval and record displacements, to the left as negative and to the right as positive.
  2. Divide the mean square by twice the interval. The answer does not change with the interval: it is D.

This is the observable the paper chooses. In §5 it predicts how far a particle gets in one second and in one minute, not how fast it moves.

05 / Ask an interval question

A curve’s height is not a probability.

“How likely is the displacement to lie between −1 and +1 micrometres?” is an interval question. The model answers with an area, a dimensionless probability. The height of the density is measured per micrometre.

At the very start every particle sits at the starting point, so any stretch that includes it holds all of them. After that, the chance of being at one exact point is zero, however crowded its neighbourhood: only a stretch of some width has a chance, and that chance is the area under the curve across it.

What happens in a more viscous liquid?

Keeping temperature and radius fixed, doubling viscosity halves the diffusion coefficient. It does not halve the RMS displacement: the square-root relationship reduces that displacement by a factor of the square root of two.

Try changing viscosity from 1 to 2 mPa·s. The laboratory does not silently change viscosity when you edit temperature; they are separate settings.

06 / Check it against the world

A number a microscope can follow.

Everything in the displacement is something a laboratory can set or read: the temperature, the viscosity of the liquid, the radius of the particle, and how long you wait. Give the ensemble below Einstein’s inputs and it gives his numbers. Give it your own and it gives theirs.

A reproducible trial

The tracer ensemble, for the check

Compare the movement of individual tracers with the statistics of the whole ensemble. Changing when or how you observe the trial does not generate different paths.

Predict before the numbers

If you watch four times as long, does the typical distance from the start become 4 times as large, 2 times as large, or unchanged?

Three relations the model could have

Predict before the numbers

In a liquid twice as viscous, after the same time, does the typical distance become half as large, about 0.71 times as large, or unchanged?

Three relations the model could have

The result appears when you choose, say you have one in mind, or skip.

These buttons use the accepted trial, not unsaved draft edits. The requested time must lie on its recording grid.

Same-seed viscosity comparisons use common random numbers, not independent trials.

Experiment settings temperature, viscosity, radius, tracers, timing, seed
Record and observe
With the same settings, the same seed replays the same trial. Any whole number from 0 to 18,446,744,073,709,551,615 works.

The preview records at most 8 MiB of latent paths. Its default is 400 tracers for 10 seconds at 0.02-second resolution. Larger requests are refused, never silently reduced.

1 μm−5 μm+5 μmy displacement
Displacements from a common origin, not a literal microscope image. Lines connect recorded positions and do not supply an instantaneous velocity. 0 / 400 endpoints lie outside this view; none are removed from the ensemble.

Static worked example

Model note
  • Temperature (temperature), Dynamic viscosity (viscosity), Particle radius (particleRadius), Observation interval (observationInterval) and Number of tracers in the trial (ensembleSize): Host calculation, by the site's own reference code. Computed in bm01.acceptedInputs.
  • Boltzmann's constant (boltzmannConstant): Host calculation, by the site's own reference code. Computed in constants.modernSI2019.
  • Diffusion coefficient (diffusionCoefficient): Host calculation, by the site's own reference code. Computed in diffusion.stokesEinsteinD.
  • RMS displacement (one dimension) (rmsDisplacement1d), RMS coordinate displacement after 1 s (lambdaX1s) and RMS coordinate displacement after 60 s (lambdaX60s): Host calculation, by the site's own reference code. Computed in diffusion.rmsDisplacement.
  • Model total mean square (modelSecondMoment), Model mean distance (modelMeanNorm) and Model RMS distance (modelRmsNorm): Host calculation, by the site's own reference code. Computed in diffusion.moments.
  • Tracer positions at the end of the interval (tracerPositions), Coordinates along the drawn traces (traceCoordinates) and Random draws in the recording (recordingDraws): Host calculation, by the site's own reference code. Computed in diffusion.recordTracers.
  • Model apparent coordinate speed (modelApparentSpeed): Host calculation, by the site's own reference code. Computed in diffusion.apparentSpeed.
  • Sample apparent coordinate speed (sampledApparentSpeed), Signed coordinate mean, compared with its sampling band (signedMean) and Mean-square coordinate displacement, compared with its sampling band (meanSquare): Host calculation, by the site's own reference code. Computed in diffusion.ensembleMoments.
  • Times along the drawn traces (traceTimes), Times on the comparison plot (plotTimes), Sample signed mean at each plotted time (plotSampleMean), Sample total mean square at each plotted time (plotSampleMsd), Sample RMS distance at each plotted time (plotSampleRms), Sample apparent speed at each plotted time (plotSampleApparent), Model signed mean at each plotted time (plotModelMean), Model total mean square at each plotted time (plotModelMsd), Model RMS distance at each plotted time (plotModelRms), Model apparent speed at each plotted time (plotModelApparent) and Whether the existing recording was reused (reusedRecording): Host calculation, by the site's own reference code. Computed in bm01.measure.
  • Edges of the histogram bins (histogramEdges), Count in each histogram bin (histogramCounts), Share of the sample in each histogram bin (histogramFrequencies), Samples below the plotted range (underflow), Samples above the plotted range (overflow) and Kolmogorov distance (kolmogorovDistance): Host calculation, by the site's own reference code. Computed in diffusion.displacementHistogram.
  • Model probability of each histogram bin (histogramModel): Host calculation, by the site's own reference code. Computed in diffusion.intervalProbability.
  • Sampling band for the signed coordinate mean (meanBand), Sampling band for the total mean square (secondMomentBand), Lower edge of the sampling band for the signed mean (signedMeanLowerBand), Upper edge of the sampling band for the signed mean (signedMeanUpperBand), Lower edge of the sampling band for the mean square (meanSquareLowerBand) and Upper edge of the sampling band for the mean square (meanSquareUpperBand): Host calculation, by the site's own reference code. Computed in diffusion.ensembleMomentBands.
  • Sample signed mean (sampleMean), Mean absolute coordinate displacement (sampleMeanAbsolute), Sample mean square (sampleMeanSquare), Sample coordinate RMS (sampleRms), Sample mean distance (sampleMeanNorm), Sample total mean square (sampleMeanSquareNorm) and Sample RMS distance (sampleRmsNorm) (secondary results): Host calculation, summarizing the tracer positions above. Computed in diffusion.ensembleMoments.
  • Seed 1905: with the same settings, the same seed replays the same trial.
  • Computed from the settings last applied (settings revision 1, result 1).
  • Not modeled: Molecular collisions (no collision bath owns the displacement).
  • Show the code

Accepted synthetic trial: 400 tracers, 1 second, coordinate mean 0.0029237 micrometres and coordinate RMS 0.98602 micrometres.

Accepted trial: seed 1905; 400 tracers; 293.15 K; viscosity 1 mPa·s; radius 0.5 μm. Observe at 1 s in a 10-second recording. Constants: modern SI 2019.

t = 1.00 s, true rate (1 s/s)
At the natural rate a tracer moves about 0.93 μm along one axis in 1 s (root mean square; scale bar: 1 μm)Simulation view: accelerated snapshot across 10 s
View only: no calculation or new draws.
Fraction in each binChoose an answer aboveto see the result−4.6338 μm4.6338 μm
This plot appears when you choose an answer above, say you have one in mind, or skip.Solid bars: the synthetic sample. Dashed line: probabilities of the same bins under the unbounded model, not a density curve. Counts beyond the plotted range: 0 left, 0 right.
Whole-ensemble statistics: signed coordinate x, total over 1 coordinate
Diffusion coefficient (model)0.42944 μm²/s
Signed mean (sample)0.0029237 μm (modern SI 2019)
Mean absolute coordinate displacement0.77808 μm (modern SI 2019)
Mean-square coordinate displacement0.97223 μm²
Coordinate RMS (sample / model)0.98602 / 0.92676 μm (modern SI 2019)
Mean distance (sample / model)0.77808 / 0.73945 μm (modern SI 2019)
Total mean square (sample / model)0.97223 / 0.85888 μm²
Apparent coordinate speed (sample / model)0.98602 / 0.92676 μm/s
Sampling bands under the model

99.9% per comparison, not a simultaneous guarantee across the table or repeated observations. These ranges are calculated from the model variance, not estimated from the observed sample, and are not uncertainties of the model itself.

Signed coordinate mean (μm): −0.15248 to 0.15248

Total mean square (μm²): 0.67299 to 1.0729

Hundreds of particles each take their own random path from the same start. Their average position stays near the start, but their typical distance from it keeps growing, as the square root of the time.

Section 4 assumes that each particle moves independently of every other, and that one particle's moves in successive short intervals τ are independent of each other, a displacement Δ being as likely as −Δ. From these assumptions the density of particles obeys a diffusion equation, and for particles that all start at x = 0 its solution is a Gaussian with mean zero and mean square 2Dt. The typical displacement along one axis is therefore λx = √(2Dt). It grows as the square root of the time, which is why four times the observation interval doubles the spread. The instrument draws such an ensemble and sets its sample mean and RMS beside the model's.

Modern qualifications

Einstein printed λx = √(2Dt) in Section 4 and, putting in his Section 3 result D = (RT/N) · 1/(6πkP), λx = √t · √((RT/N) · 1/(3πkP)) in Section 5. In his notation k is the fluid's viscosity, not Boltzmann's constant, and P is the particle's radius. He did not claim to have explained an observed motion: the paper opens by saying these motions may be the Brownian motion, but that the reports available to him were too imprecise to judge. Langevin reached the same mean square in 1908 from an equation of motion with a random force, and Perrin's measurements of 1908 and 1909 used Einstein's formula to estimate N.

Inspect the model behind this trial

These equations describe the same accepted trial as the plots and table. Select a symbol or an operation to see what it means; model predictions are not sample estimates.

Explore the equation · Modern model notation

Why the apparent speed depends on how you watch

vapp:=λxtv_{\mathrm{app}} := \frac{\lambda_x}{t}

Shown in modern letters; Einstein's are in §4 of the German source face.

divides

Select a term or operation. In the formula, Down enters an operation, Up returns to its parent, and Left/Right move between siblings. Escape clears selection. Tab leaves the formula.

Each term and operation, in words

Static worked example

These values are for the settings last applied, not for edits you have not applied yet.

Apparent coordinate speed
0.92676 μm/s
Coordinate RMS displacement
0.92676 μm
Observation interval
1 s
Read the equation aloud in words

Apparent coordinate speed equals the model coordinate root mean square displacement divided by the observation interval.

This quotient depends on the observation interval. It is not instantaneous physical velocity. At zero interval the quotient is undefined, even though the displacement is zero.

Model assumptions and every term’s meaning
  • The same one-coordinate ideal Brownian model as the RMS relation.
  • A positive observation interval is required for the quotient.
  • Lines joining recorded points are a rendering convention, not a velocity measurement.
Define the observable
This defines an interval-dependent comparison. It does not introduce a physical instantaneous Brownian velocity. Read the prerequisite
Apparent coordinate speed
A distance-per-interval statistic. It is not a molecular collision speed. Read the prerequisite
Divide by the same interval
The distance grows as the square root of time, while the denominator grows linearly. The quotient therefore decreases as the interval grows. Read the prerequisite
Model coordinate RMS
The same accepted one-coordinate model displacement used by the neighboring RMS equation. Read the prerequisite
Positive observation interval
A zero interval yields no apparent-speed value. The interface keeps the explanation instead of fabricating zero or infinity. Read the prerequisite

Every term’s units were checked when this page was built. That checks the units, not the model. This equation is written for this edition in modern notation, not transcribed from the paper.

Explore the equation · Modern model notation

Resistance to motion controls spreading

D=kB T6 π η aD = \frac{k_B\,T}{6\,\pi\,\eta\,a}

Shown in modern letters; Einstein's are in §3 of the German source face.

is thermal energy divided by the viscous drag coefficient.

Select a term or operation. In the formula, Down enters an operation, Up returns to its parent, and Left/Right move between siblings. Escape clears selection. Tab leaves the formula.

Each term and operation, in words

Static worked example

These values are for the settings last applied, not for edits you have not applied yet.

Diffusion coefficient
0.42944 μm²/s
Boltzmann constant
1.3806 × 10−23 J/K
Absolute temperature
293.15 K
Dynamic viscosity
1 mPa·s
Particle radius
0.5 μm
Read the equation aloud in words

The diffusion coefficient equals the Boltzmann constant times the absolute temperature, divided by six times pi times the dynamic viscosity times the particle radius.

Thermal energy competes with viscous drag. Doubling viscosity halves the diffusion coefficient while reducing RMS displacement only by the square root of two. Radius means radius, not diameter.

Model assumptions and every term’s meaning
  • Dilute, approximately spherical tracers with no-slip Stokes drag in a homogeneous Newtonian liquid.
  • Wall corrections, interactions, inertia, and observation noise are not included.
  • The modern SI 2019 constant set is used; this is not a historical inversion exercise.
An ideal model
This equation assumes the dilute spherical tracer and Stokes-drag model; dimensional consistency alone does not establish those assumptions. Read the prerequisite
Diffusion coefficient
A squared-distance-per-time coefficient. This value comes from the accepted model calculation. Read the prerequisite
Why divide by drag?
Increasing the denominator reduces diffusivity at fixed thermal energy. This is a model dependence, not a rule that every larger quantity must reduce motion. Read the prerequisite
Thermal energy scale
Boltzmann constant times absolute temperature supplies energy per particle. Read the prerequisite
A known modern constant
The SI 2019 constant set supplies this value. It must not be treated as independent historical evidence when trying to infer molecular number. Read the prerequisite
Absolute temperature
Use kelvin. The laboratory holds viscosity as an independently supplied parameter. Read the prerequisite
The drag coefficient
Six pi times viscosity times radius is Stokes drag per speed for an isolated sphere in the assumed low-Reynolds regime. Read the prerequisite
Dynamic viscosity
Viscosity is in the denominator. Doubling it at fixed temperature and radius halves D, not the RMS displacement. Read the prerequisite
Particle radius
The input is a radius, not a diameter. Confusing them changes the predicted coefficient by a factor of two. Read the prerequisite

Every term’s units were checked when this page was built. That checks the units, not the model. This equation is written for this edition in modern notation, not transcribed from the paper.

Explore the equation · Modern model notation

From spreading to a measurable distance

λx=2 D t\lambda_x = \sqrt{2\,D\,t}

Shown in modern letters; Einstein's are in §4 of the German source face.

is the positive square root

Select a term or operation. In the formula, Down enters an operation, Up returns to its parent, and Left/Right move between siblings. Escape clears selection. Tab leaves the formula.

Each term and operation, in words

Static worked example

These values are for the settings last applied, not for edits you have not applied yet.

Coordinate RMS displacement
0.92676 μm
Diffusion coefficient
0.42944 μm²/s
Observation interval
1 s
Read the equation aloud in words

The model coordinate root mean square displacement equals the square root of two times the diffusion coefficient times the observation interval.

Squaring measures spread without cancellation between directions. Taking the positive square root turns squared distance back into a distance. Four times the observation interval gives twice the model RMS, not four times.

Model assumptions and every term’s meaning
  • Independent, zero-mean Gaussian displacement increments in a homogeneous liquid.
  • One-coordinate model statistic, not a measured speed or a sample estimate.
  • The overdamped regime is assumed rather than established from additional particle and fluid measurements.
A model relation
This equality concerns the ideal model. A finite synthetic ensemble fluctuates around it; its sample RMS is displayed separately. Read the prerequisite
Coordinate RMS
One coordinate, not the total three-dimensional distance. Model RMS and sample RMS have different meanings. Read the prerequisite
Why a square root?
Two D t has units of squared length. Its positive square root has units of length and defines the typical displacement. Read the prerequisite
Build the mean square
Independent zero-mean increments add their variances. The definition of D makes the coordinate mean square equal to 2 D t. Read the prerequisite
Diffusion coefficient
This is the accepted model diffusivity, not a rate inferred from the synthetic data. Read the prerequisite
Observation interval
This is the interval used for the displayed displacement, not the simulation frame rate. Re-observing the trial preserves its paths. Read the prerequisite

Every term’s units were checked when this page was built. That checks the units, not the model. This equation is written for this edition in modern notation, not transcribed from the paper.

RMS distance over recorded time

Choose an answer aboveto see the resultμm · log scale0.02 s10 s
This plot appears when you choose an answer above, say you have one in mind, or skip.Solid: this sample. Dashed: the model. The time axis is logarithmic. Zero sample values, if any, remain in the table but cannot appear on a log scale. All times refer to the same recorded paths.
Read the comparison as a table
RMS distance (μm)
Time (s)SampleModel
0.020.130380.13106
0.040.185560.18535
0.10.303680.29307
0.20.404160.41446
0.40.613880.58613
10.986020.92676
21.34951.3106
41.90351.8535
102.87232.9307

What is, and is not, being simulated

These synthetic paths are Gaussian independent increments for dilute spherical tracers in a homogeneous Newtonian liquid. They do not simulate individual molecular collisions, inertia, interactions, sedimentation, walls, localization error or motion blur.

Low Reynolds number and observation times long compared with momentum relaxation are assumed, not verified from fluid and particle data. Lines between samples are a drawing convention; they do not define an instantaneous Brownian speed.

The three latent coordinates are recorded together. Axis, dimension, observation and statistic changes reuse those coordinates. The histogram and statistics use every tracer, including those outside the view.

No FrankenSim WASM artifact produced this trial. Integer random draws follow the pinned Philox mapping; Gaussian conversion uses this host’s math functions and is not claimed bitwise identical across browser engines.

Logical recording draws: 1200000. This result was assembled from a newly generated deterministic recording.

Evaluator source digest: source:sha256:6e9e4cb91ff4a2493d2b641284d3697eec06b64a62f056de23ddc501ecb7aa67

Show the code

The site’s own reference code for this law, in TypeScript. It computes the numbers above when the page runs it, and it is what the page uses wherever a FrankenSim result is not available.

stokesEinsteinD · src/physics/reference/diffusion/distributions.ts · this site’s source, as built · sha256:a41fa1ff39026dbed9a701b368028012995d1835d9acebb2e32d5010f47404ec

This is the function that produced the numbers shown now.

In words

Take the temperature and the gas constant, divide by the number of molecules in a mole to get the energy scale of one molecule, then divide by the drag on a sphere of this radius in a liquid of this viscosity.

Mathematics

Resistance to motion controls spreading

D=kB T6 π η aD = \frac{k_B\,T}{6\,\pi\,\eta\,a}

Explore Resistance to motion controls spreading term by term, with this laboratory’s values

Implementation

/** SI: temperature K, viscosity Pa s, radius m, output m2/s. No ambient constants. */
export function stokesEinsteinD(
  {
    T,
    eta,
    a,
    medium = "liquid",
  }: { T: number; eta: number; a: number; medium?: "liquid" | "gas" },
  set: ConstantSet,
): Evaluation {
  if (![T, eta, a].every((v) => Number.isFinite(v) && v > 0))
    return outside(
      "diffusionCoefficient",
      "stokesEinsteinD",
      "T > 0, eta > 0, a > 0",
      "Enter positive finite temperature, viscosity, and particle radius.",
      "input",
      set,
    );
  if (medium !== "liquid")
    return outside(
      "diffusionCoefficient",
      "stokesEinsteinD",
      "stokes-gas-medium",
      "The liquid Stokes-drag model does not include the slip correction needed in a gas.",
      "model",
      set,
    );
  const k = thermalConstant(set);
  return number(
    "diffusionCoefficient",
    "stokesEinsteinD",
    (k.value * T) / (6 * Math.PI * eta * a),
    set,
    undefined,
    true,
  );
}
One worked example of this calculation. Constant set: Declared 1905-plan inputs.
StepExpressionValueUnit
The gas constant in SIR = 8.31 × 107 erg mol−1 K−18.31J mol⁻¹ K⁻¹
Thermal energy per moleRT2411.1J mol⁻¹
Drag on one sphere6 π η a (this operation in the equation)1.2723 × 10⁻⁸kg s⁻¹
Drag on a mole of spheresN 6 π η a7.6341 × 10¹⁵kg s⁻¹ mol⁻¹
Diffusion coefficientRT / (N 6 π η a) (this operation in the equation)3.1584 × 10⁻¹³m² s⁻¹
Coordinate RMS at 1 sλ_x = √(2 D t) at t = 1 s (this operation in the equation)7.9478 × 10⁻⁷m
Coordinate RMS at 60 sλ_x = √(2 D t) at t = 60 s (this operation in the equation)6.1564 × 10⁻⁶m

The site’s own reference code for this law, in TypeScript. It computes the numbers above when the page runs it, and it is what the page uses wherever a FrankenSim result is not available.

rmsDisplacement · src/physics/reference/diffusion/distributions.ts · this site’s source, as built · sha256:6d2d381222179a43397683f83d95eecf9dab253844f72dfcaa1eaab1302222eb

This function computes the listed outputs when it runs.

In words

The typical one-dimensional displacement is the square root of twice the diffusion coefficient times the observation interval.

Mathematics

From spreading to a measurable distance

λx=2 D t\lambda_x = \sqrt{2\,D\,t}

Explore From spreading to a measurable distance term by term, with this laboratory’s values

Implementation

export function rmsDisplacement(D: number, t: number): Evaluation {
  if (!validDt(D, t))
    return outside(
      "rmsDisplacement1d",
      "rmsDisplacement",
      "D >= 0 and t >= 0",
      "Diffusivity and elapsed time must be finite and nonnegative.",
    );
  return number(
    "rmsDisplacement1d",
    "rmsDisplacement",
    scale(D, t),
    undefined,
    undefined,
    D > 0 && t > 0,
  );
}

The site’s own reference code for this law, in TypeScript. It computes the numbers above when the page runs it, and it is what the page uses wherever a FrankenSim result is not available.

apparentSpeed · src/physics/reference/diffusion/distributions.ts · this site’s source, as built · sha256:1f2137435fdc15b0a9b5c91cc4dafba680e8d52b0f5143fdbefdfa9945f6baec

This function computes the listed outputs when it runs.

In words

Divide the model RMS displacement by the observation interval. The quotient depends on how long you watch.

Mathematics

Why the apparent speed depends on how you watch

vapp:=λxtv_{\mathrm{app}} := \frac{\lambda_x}{t}

Explore Why the apparent speed depends on how you watch term by term, with this laboratory’s values

Implementation

export function apparentSpeed(D: number, tau: number): Evaluation {
  if (!validDt(D, tau) || tau === 0)
    return outside(
      "apparentSpeed",
      "apparentSpeed",
      "D >= 0 and tau > 0",
      "An apparent speed needs a positive observation interval.",
    );
  return number(
    "apparentSpeed",
    "apparentSpeed",
    (Math.SQRT2 * Math.sqrt(D)) / Math.sqrt(tau),
    undefined,
    undefined,
    D > 0,
  );
}

The site’s own reference code for this law, in TypeScript. It computes the numbers above when the page runs it, and it is what the page uses wherever a FrankenSim result is not available.

ensembleMoments · src/physics/reference/diffusion/tracers.ts · this site’s source, as built · sha256:1d87c8e69590b8871bcc7a6390089003f59d0b80bfc7eb1919a57dec661743e6

This function computes the listed outputs when it runs.

In words

Average the recorded displacements of every tracer, not a viewport subset.

Implementation

/** Compensated sums over ALL members, never a viewport-selected subset. Row-major M by d. */
export function ensembleMoments({
  displacements,
  d,
  dt,
  viewport,
}: {
  displacements: Float64Array;
  d: number;
  dt?: number;
  viewport?: ViewportBounds;
}): Computation<EnsembleMoments> {
  if (
    !(displacements instanceof Float64Array) ||
    ![1, 2, 3].includes(d) ||
    displacements.length === 0 ||
    displacements.length % d !== 0 ||
    displacements.length > 30000 ||
    !displacements.every(Number.isFinite)
  )
    return invalid(
      ["displacements", "d"],
      "Provide finite, row-major displacements with one, two or three coordinates per tracer.",
    );
  const M = displacements.length / d,
    sums = new Float64Array(d * 3 + 1),
    corrections = new Float64Array(d * 3 + 1);
  function add(i: number, v: number) {
    const corr = corrections[i] ?? 0,
      sum = sums[i] ?? 0,
      y = v - corr,
      t = sum + y;
    corrections[i] = t - sum - y;
    sums[i] = t;
  }
  for (let i = 0; i < M; i++) {
    let norm = 0;
    for (let j = 0; j < d; j++) {
      const v = displacements[i * d + j];
      if (v === undefined) return failure("A required displacement coordinate was undefined.");
      if (v !== 0 && v * v === 0)
        return failure("A nonzero squared displacement is below the representable range.");
      add(j * 3, v);
      add(j * 3 + 1, Math.abs(v));
      add(j * 3 + 2, v * v);
      norm = Math.hypot(norm, v);
    }
    add(d * 3, norm);
  }
  const axes = Array.from({ length: d }, (_, j) => {
    const mean = (sums[j * 3] ?? 0) / M;
    const meanAbsolute = (sums[j * 3 + 1] ?? 0) / M;
    const meanSquare = (sums[j * 3 + 2] ?? 0) / M;
    return {
      mean,
      meanAbsolute,
      meanSquare,
      rms: Math.sqrt(meanSquare),
    };
  });
  const meanSquareNorm = axes.reduce((s, a) => s + a.meanSquare, 0);
  if (![...sums, meanSquareNorm].every(Number.isFinite))
    return failure("The moment reduction exceeded the numerical range.");
  const rmsNorm = Math.sqrt(meanSquareNorm);
  let insideCount: number | undefined;
  let outsideCount: number | undefined;
  if (viewport) {
    let inc = 0;
    let outc = 0;
    for (let i = 0; i < M; i++) {
      let isInside = true;
      for (let j = 0; j < d; j++) {
        const v = displacements[i * d + j] ?? 0;
        const minVal = viewport.min[j] ?? -Infinity;
        const maxVal = viewport.max[j] ?? Infinity;
        if (v < minVal || v > maxVal) {
          isInside = false;
          break;
        }
      }
      if (isInside) inc++;
      else outc++;
    }
    insideCount = inc;
    outsideCount = outc;
  }
  const apparentSpeed =
    dt !== undefined && Number.isFinite(dt) && dt > 0
      ? (d === 1 ? (axes[0]?.rms ?? 0) : rmsNorm) / dt
      : undefined;
  return {
    kind: "accepted",
    data: {
      M,
      axes,
      meanNorm: (sums[d * 3] ?? 0) / M,
      meanSquareNorm,
      rmsNorm,
      ...(apparentSpeed !== undefined ? { apparentSpeed } : {}),
      ...(insideCount !== undefined && outsideCount !== undefined ? { insideCount, outsideCount } : {}),
    },
  };
}

The site’s own reference code for this law, in TypeScript. It computes the numbers above when the page runs it, and it is what the page uses wherever a FrankenSim result is not available.

displacementHistogram · src/physics/reference/diffusion/tracers.ts · this site’s source, as built · sha256:746f24cee7b8573b8a986cd8e072b4786549eba47dffca106dbd38582d670c7b

This function computes the listed outputs when it runs.

In words

Count how many recorded displacements fall in each interval of the histogram.

Implementation

export function displacementHistogram(
  values: Float64Array,
  edges: Float64Array,
): Computation<{ counts: Float64Array; underflow: number; overflow: number; total: number }> {
  if (
    values.length === 0 ||
    values.length > 10000 ||
    edges.length < 2 ||
    edges.length > 1001 ||
    !values.every(Number.isFinite) ||
    !edges.every((v, i) => {
      if (!Number.isFinite(v)) return false;
      if (i === 0) return true;
      const prev = edges[i - 1];
      return prev !== undefined && v > prev;
    })
  )
    return invalid(
      ["values", "edges"],
      "Use finite samples and strictly increasing histogram edges.",
    );
  const counts = new Float64Array(edges.length - 1);
  let underflow = 0,
    overflow = 0;
  const firstEdge = edges[0];
  const lastEdge = edges[edges.length - 1];
  if (firstEdge === undefined || lastEdge === undefined)
    return invalid(["edges"], "Histogram edges array must contain at least two finite bounds.");
  for (const value of values) {
    if (value < firstEdge) {
      underflow++;
      continue;
    }
    if (value > lastEdge) {
      overflow++;
      continue;
    }
    let lo = 0,
      hi = edges.length - 1;
    while (hi - lo > 1) {
      const mid = (lo + hi) >>> 1;
      const edgeMid = edges[mid];
      if (edgeMid !== undefined && value < edgeMid) hi = mid;
      else lo = mid;
    }
    const bin = Math.min(lo, counts.length - 1);
    const prevCount = counts[bin];
    if (prevCount !== undefined) {
      counts[bin] = prevCount + 1;
    }
  }
  return { kind: "accepted", data: { counts, underflow, overflow, total: values.length } };
}
Every histogram count, including tails
Coordinate intervals in μm; half-open bins, with the final endpoint included
IntervalCount
Below the plotted range0
−4.6338 to −4.40210
−4.4021 to −4.17040
−4.1704 to −3.93870
−3.9387 to −3.7070
−3.707 to −3.47530
−3.4753 to −3.24370
−3.2437 to −3.0120
−3.012 to −2.78031
−2.7803 to −2.54861
−2.5486 to −2.31690
−2.3169 to −2.08527
−2.0852 to −1.85357
−1.8535 to −1.62186
−1.6218 to −1.390116
−1.3901 to −1.158412
−1.1584 to −0.9267615
−0.92676 to −0.6950728
−0.69507 to −0.4633824
−0.46338 to −0.2316939
−0.23169 to 039
0 to 0.2316943
0.23169 to 0.4633837
0.46338 to 0.6950726
0.69507 to 0.9267632
0.92676 to 1.158423
1.1584 to 1.390117
1.3901 to 1.62188
1.6218 to 1.85356
1.8535 to 2.08526
2.0852 to 2.31693
2.3169 to 2.54861
2.5486 to 2.78031
2.7803 to 3.0121
3.012 to 3.24371
3.2437 to 3.47530
3.4753 to 3.7070
3.707 to 3.93870
3.9387 to 4.17040
4.1704 to 4.40210
4.4021 to 4.63380
Above the plotted range0
Total, including both tails400

Check it against the world

A printed prediction

Einstein predicted a displacement a microscope could follow: for particles 0.001 mm across in water at 17 °C, about 0.8 micron in one second and about 6 microns in one minute. Set the tracer ensemble to his inputs, or to your own, and compare.

Worked value

What Einstein printed in §5, taking N = 6 × 10²³ from gas theory

about 0.8 µm in one second, about 6 µm in one minute

From the tracer ensemble above, as it now stands

At 20 °C, a viscosity of 1 mPa·s and a radius of 0.5 µm, a particle typically gets:

In one second
0.93 µm
In one minute
7.2 µm

An ideal model computed here, with today’s constants. Change the radius in the ensemble and apply it, and these numbers follow.

Later evidence, not on the 1904 shelf

Published, 1909-09

Perrin measures suspensions of spherical gamboge and mastic grains of measured radius, among them their vertical distribution in sedimentation equilibrium, and infers N, the number of molecules in a gram-molecule. His three Brownian-motion methods give 70.5, 71.5 and 65 × 10²² (the distribution of a uniform emulsion, the mean displacement, the mean rotation), and he takes 70.5 × 10²², about 7 × 10²³, as the most probable value.

Later confirmation
Limits: Later evidence, not on the 1904 shelf. The same table sets these values beside those of other phenomena, from the viscosity of gases to the infra-red spectrum, which range from 30 to 200 × 10²². The numbers here are read from that table; the dataset behind them is not transcribed.

Historical context

Later empirical or theoretical confirmation developed after 1905.

Available by

Published, 1909-09

What the paper itself cites or asserts

Not yet recorded for this card.

Where this site uses it

Primary evidence and sources

  • Mouvement brownien et réalité moléculaire J. Perrin, Ann. Chim. Phys. (8) 18 (1909) 5 (1909-09)

The later evidence is not a displacement for these particles. Perrin measured suspensions of his own, and what he reported was the number of molecules that makes the formulas fit them. It came out close to the 6 × 1023 Einstein took from gas theory, and that number is what sets the size of the displacements he printed.

07 / Turn the question around

What can a finite sample tell you?

In the next laboratory, the molecular number behind a made-up set of paths is hidden. How fast the particles spread fixes the radius and the number only together: a larger radius with a smaller number spreads at the same rate. So give it the radius from a separate measurement, estimate the number, and run it again and again to see how often the range you quote misses the true number.

08 / Try it yourself

From one axis to two

A microscope sees the plane, not one axis. Work out how far a particle gets in two dimensions from what one axis gives you. Any correct form is accepted: the checker compares your expression with the answer at sample values of D and t across the tracer laboratory's ranges, not by matching text.

Along one axis the mean square displacement is 2·D·t. Watch the same particle in the plane, moving along x and y independently. What is its root mean square distance from the start?

Answers are compared for D from 1 × 10−14 to 1 × 10−10 and t from 0.1 to 100.

Show a worked explanation

The squared distance from the start is x² + y². Each axis contributes a mean square of 2·D·t, independently, so the mean square distance is 4·D·t, and its root is sqrt(4·D·t) = 2·sqrt(D·t). That is sqrt(2) times the one-axis value, sqrt(2·D·t), not twice it: the two mean squares add, not the two distances.

Now the number itself. With Einstein’s own inputs the formula gives a distance a laboratory could go and measure. Type it in the unit you prefer.

Einstein put in numbers of his own. Water at 17 °C, whose viscosity he took as 1.35 × 10⁻³ Pa·s; particles 0.001 mm across; R = 8.31 J/(mol·K) and N = 6 × 10²³ per mole. How far does such a particle typically move along one axis in one second?

Show a worked explanation

First the diffusion coefficient, D = RT/(N · 6πηa). The formula needs the radius, and 0.001 mm across is a radius of 0.5 μm, which is 5 × 10⁻⁷ m. With T = 290.15 K, D = (8.31 × 290.15)/(6 × 10²³ × 6π × 1.35 × 10⁻³ × 5 × 10⁻⁷), about 3.16 × 10⁻¹³ m²/s. Along one axis the mean square displacement is 2·D·t, so after one second the typical distance is √(2 × 3.16 × 10⁻¹³ × 1) m, about 7.95 × 10⁻⁷ m, or 0.795 μm. Einstein printed it as 0.8 micron.

The inputs carry two or three significant figures, so an answer within 1 percent is the same answer; Einstein's own 0.8 micron is accepted.

Predict, change one thing, then explain

  1. Predict

    Predict first. The tracer ensemble shows how far a particle typically gets in one second. If you double the particle's radius, does that distance halve, fall by less than half, or stay the same?

  2. Change one thing

    Open the tracer ensemble, change the radius from 0.5 to 1 micrometre, apply it, and read the one-second displacement again.

  3. Explain

    Explain the factor you saw. Say what the radius does to the drag, what the drag does to D, and how the distance depends on D.

Where this enters the paper

The Brownian-motion paper’s §§4–5 connect irregular displacements to diffusion and then to a measurable displacement scale. The German text and an English translation are on this site. This route is new explanation written for this edition, not a translation of the paper.

In the 1905 paper

The second route, force against drag, is the paper's §3. Its result is the diffusion coefficient that §5 turns into a displacement.

Read the argument: what fixes D for a small sphere →

In the 1905 paper

The move is in §4, where a particle's displacements over successive intervals are taken to be independent of each other.

Read the argument: why independent steps add →

In the 1905 paper

The paper closes §5 by turning the relation around: measure the displacement, and the one unknown left is the number of molecules.

Read the argument: what would let us count molecules →

Two doors lead to the same coefficient: the paper’s balance of a crowd, and one particle’s walk followed step by step. The route takes both, and they meet.

Ways in

Multiple routes, one arrival point

All doors arrive at the same result: the diffusion coefficient of a suspended sphere, D = RT/(6πηaN), the paper's relation in today's letters

the paper's routeanother routeone result

The drawing gives every route the same box and the same line on purpose. Neither door is the real one and neither is a detour: the paper's own construction and the routes beside it reach the result named above, and the journey says so. What it does not show is what each route costs a reader to walk, which is the only respect in which they genuinely differ and is written in the cards below rather than in the picture.

The paper's route

The paper's balance, §§1 to 3

Grant that a visible particle presses on a membrane the way a dissolved molecule does, pull on the particles with a force, and let the crowd settle. The force cancels between the two ways of counting what crosses a plane each second, and the coefficient left behind depends on the temperature, the viscosity of the liquid and the radius of the particle alone.

Arrives at: the diffusion coefficient of a suspended sphere, D = RT/(6πηaN), the paper's relation in today's letters

Another route

Follow one particle: the walk, for programmers

Write the loop instead: many short independent steps, each as likely to go one way as the other. The mean square of the total grows in proportion to the time, and the coefficient of that growth is the same D the balance fixes, which is why the spreading a microscope can watch measures the quantity the thermodynamic argument computes.

Arrives at: the diffusion coefficient of a suspended sphere, D = RT/(6πηaN), the paper's relation in today's letters

The two doors can also be taken apart and put back together. The capstone separates what the balance of a crowd establishes from what one particle’s walk establishes, and asks which of them has to come before which: Rebuild the argument for the spread. It is optional, and nothing above depends on it.