Kagan's non-linear effect: when an impure catalyst gives a pure product

In 1986 Henri Kagan showed that a chiral catalyst's purity need not pass straight to its product. How the effect works, the models behind it, why chemists use it to read reaction mechanisms, and how to fit it to your own data.

Part 1 ended with a common-sense rule: a catalyst that is 60 % pure in one hand should give a product that is 60 % as pure as the best it can make. In 1986 Henri Kagan and his team in Orsay showed that this rule can fail, sometimes in your favour and sometimes against you. The failure turned out to be a gift: it lets chemists make very pure products from impure catalysts, and it is a window into how a catalyst actually works. This part explains the effect, the simple model that captures it, and how to fit it to your own data with our new chiral amplification lab.

In short

  • Many catalysts carry two chiral ligands. With an impure ligand, the catalyst forms matched pairs (RR or SS) and mixed pairs (RS).
  • If mixed pairs are favoured and lazy, they lock away the minor hand. The catalysts left working are purer than the ligand, so the product is purer than expected: a positive non-linear effect, or asymmetric amplification.
  • If mixed pairs are busy, they make racemic product and drag the purity down: a negative effect.
  • Kagan’s ML2 model (1986, worked out in full in 1994) predicts the curve from two numbers. Chemists now measure such curves routinely to learn how many ligand molecules sit in the working catalyst.

The expected straight line

Call the ligand’s enantiomeric excess ee_L and the best product ee (with a pure ligand) ee_max. If every catalyst molecule worked independently, an R-catalyst would make mostly R product and an S-catalyst mostly S, and the two contributions would simply add. The product’s ee would be

ee_product = ee_max × ee_L

a straight line from (0, 0) to (100 %, ee_max). For forty years after asymmetric catalysis began, everyone assumed this held.

1986: three reactions, three different answers

Kagan, with Christian Puchot, Olivier Samuel, Elisabet Duñach, Shu-Hai Zhao and Claude Agami, tested the assumption on purpose. They ran known asymmetric reactions with ligands of deliberately reduced purity and plotted product ee against ligand ee:

  • In the Sharpless epoxidation of geraniol (adding an oxygen across a double bond with a titanium catalyst built from diethyl tartrate), the points rose above the line: a positive effect.
  • In Kagan’s own sulfoxidation (a related titanium–tartrate system that turns sulfides into chiral sulfoxides), they fell below it: a negative effect.
  • A third case, the proline-catalysed aldol reaction, looked slightly negative at the time; much later, in 2003, a careful re-examination found it linear after all, a reminder that these effects must be measured with care.

The paper was titled simply “Nonlinear effects in asymmetric synthesis”, and the name stuck (often shortened to NLE, written (+)-NLE or (−)-NLE). Hans Wynberg and Ben Feringa had pointed out ten years earlier that the two hands of a molecule can interact when mixed, so a mixture is not just the sum of its parts; Kagan turned that idea into a measurable, predictable effect in catalysis.

The ML2 model: pairs of ligands

Kagan’s explanation is beautifully simple. Suppose the working catalyst is a metal M carrying two ligands L. With an impure ligand three kinds of complex form:

  • ML_R L_R (call its share x) and ML_S L_S (share y): the homochiral pairs, mirror images of each other, equally good at making their own hand of product;
  • ML_R L_S (share z): the heterochiral, “meso” pair, which is not chiral overall and makes racemic product (if it reacts at all).

The pairs swap ligands until they reach an equilibrium with constant K = z² / (x·y). Pure chance gives K = 4 (just as tossing two coins gives one head and one tail twice as often as two heads). Larger K means mixed pairs are favoured. Let g be how fast a mixed pair reacts compared with a matched pair. Then the product’s ee is

ee_product = ee_max × ee_L × (1 + β) / (1 + g β), with β = z / (x + y)

If g = 1 the fraction is 1 and the line is straight. If g < 1 the mixed pairs are slow and the curve bows up; if g > 1 it sags down. Play with it below: the dumbbells are the pairs, the curve is the model.

Interactive model: Kagan's ML2 non-linear effect (needs JavaScript).

Solving the model, and its bigger cousins

With x + y + z = 1 and x − y = ee_L (each RR pair carries two R ligands, each RS one of each), K = z²/(xy) becomes a quadratic in z: (4 − K) z² + 2K z − K (1 − ee_L²) = 0. Pick the root between 0 and 1 − ee_L. The rate of the reaction relative to a pure catalyst is x + y + g z, which is why a strong positive effect always costs speed: most of the metal is parked in the idle meso pairs, a “reservoir” that holds back the minor hand. In the MLn models (three or more ligands per metal, or catalysts that cluster into bigger aggregates) the curves get more complicated and can even cross the line. A reservoir can also sit outside the catalytic cycle altogether, for example in an insoluble racemic crystal of the ligand. Donna Blackmond showed that measuring the reaction rate together with the ee pins down which picture applies: in the ML2 reservoir picture the rate falls as the ligand gets less pure, while in some other cases it does not.

1989: amplification you can use

The most spectacular positive effect came three years later from Ryoji Noyori’s group, in the addition of diethylzinc to benzaldehyde, catalysed by a chiral amino alcohol called DAIB. With DAIB only about 15 % ee, the product alcohol formed with about 95 % ee: nearly as good as with pure DAIB. The reason fits Kagan’s picture exactly. The zinc–DAIB complexes pair up into dimers. The mixed RS dimer is much more stable than the matched ones and does not react, so it soaks up the minor hand; the small excess of matched dimers left over does all the work.

This is called asymmetric amplification, and it is useful: a cheap, impure ligand can give an almost pure product. It is also the bridge to part 3, because Kenso Soai’s self-copying reaction is the same kind of zinc chemistry, with the product itself as the catalyst.

A tool for reading mechanisms

Today measuring a non-linear effect is a standard experiment when chemists study a new asymmetric catalyst. If the curve is straight, the working catalyst probably carries one chiral ligand (or its ligands act independently). If it bends, at least two ligands meet somewhere in or near the cycle: in the active catalyst, in a resting state, or in an off-cycle reservoir. The size and shape of the bend, together with the rate, separate these cases. Thousands of papers have reported such curves since 1986, in metal catalysis, organocatalysis and enzymes alike. Related ideas grew from it: “chiral poisoning”, where a cheap chiral additive deactivates one hand of a racemic catalyst, and “asymmetric activation”, where an additive switches on one hand.

Checked with the tools

1. The three kinds of curve. The chiral amplification lab draws the ML2 curves. With K = 100 and g = 0 (favoured, idle mixed pairs) a ligand only 20 % pure gives about 80 % product ee, the kind of amplification Noyori saw. With K = 100 and g = 3 (busy mixed pairs) the curve sags well below the line.

Kagan's ML2 model: product ee against ligand eeKagan's ML2 model: product ee against ligand ee020406080100020406080100Ligand ee (%)Product ee (%)Positive: K = 100, g = 0Linear: g = 1Negative: K = 100, g = 3fibub.com
Chiral amplification lab maths (Kagan's ML2 model, ee_max = 100 %). Our own curves for round parameter values.

2. Does the formula really follow from the chemistry? Our own simulation. The ML2 formula assumes the pairs reach equilibrium. We checked that independently: starting from complexes formed at random from an impure ligand pool, we ran the ligand exchange RR + SS ⇌ 2 RS (equilibrium constant 100) to equilibrium with the adaptive Runge-Kutta method of the ODE solver, then read off the product ee. The simulated points land on the formula’s curve to within 10⁻¹⁰. The dashed line is the price: at 20 % ligand ee the catalyst runs at only 25 % of the speed of a pure one.

Checking Kagan's formula with a full simulationChecking Kagan's formula with a full simulation020406080100020406080100Ligand ee (%)Product ee (%)ML2 formula (K = 100, g = 0)ODE simulation of ligand exchangeRelative catalyst speedfibub.com
Dots: the exchange RR + SS ⇌ 2 RS run to equilibrium with the ODE solver's adaptive Runge-Kutta (RK45), starting from complexes formed at random; they match the closed formula to within 7e-11. Dashed: the price of the effect, the reaction rate relative to a pure catalyst.

3. Fitting real-looking data. The lab’s example data (made up for practice: 0→0, 10→25, 20→44, 40→71, 60→86, 80→94, 100→98, in %) fit the ML2 model with K ≈ 32 and g ≈ 0.18 (R² = 0.9998) using Levenberg-Marquardt least squares, the same method as the curve fitter.

Case Reported Our check
Matched and mixed pairs at chance (K = 4, g = 1) straight line expected straight line, to rounding
Amplification with idle mixed pairs DAIB: about 15 % ee ligand gave about 95 % ee product (Noyori, 1989) K = 100, g = 0: 20 % ligand gives 80 % product
Formula against a full exchange simulation agree to 10⁻¹⁰

How we made the graphs

scripts/blog-figures/nobel-chemistry-2026.mjs calls ml2() from src/lib/sci/chiralamp.ts (the Chiral amplification lab’s model) and rk45() from src/lib/sci/ode.ts (the ODE solver’s adaptive Dormand–Prince method). The exchange simulation is written as three rate equations, x′ = −K x y + z², y′ = −K x y + z², z′ = 2 (K x y − z²), run to time 50, by which point it has settled. No paper data are used; the parameters are round numbers.

How we test the maths

tests/sci.mjs pins the model to cases you can work by hand: a racemic ligand at K = 4 gives RR : SS : RS = 1 : 1 : 2; g = 1 gives a straight line for any K; at ee_L = 50 %, K = 4 and g = 0 the product ee is exactly 80 % (β = 0.6, so 0.5 × 1.6); the equilibrium K = z²/(xy) is recovered from the solved shares; and g above 1 gives a negative effect. The ODE solver is checked in tests/sci-tools/ode-solver.mjs against SciPy step for step: it takes the same RK45 steps, ends at the same values and matches the Lorenz and Lotka–Volterra systems to about 10⁻⁷. All run on every change.

Check it yourself

  1. Open the chiral amplification lab. It starts in Non-linear effect (Kagan) with example data in the box.
  2. Read the headline: about 44 % product ee from a 20 % pure ligand, and the fitted K and g in the facts below it.
  3. Clear the box. Set Mixed-pair constant K to 4 and Mixed-pair reactivity g to 1: the curve lies on the dashed straight line (“about linear”).
  4. Set K = 100 and g = 0: a strong positive effect. Look at Relative rate: the catalyst slows down as the curve bends up.
  5. Set g = 3: a negative effect, below the line.
  6. To see the equilibrium yourself, open the ODE solver and type x' = -K*x*y + z^2, y' = -K*x*y + z^2, z' = 2*(K*x*y - z^2) with K = 100. Start from x = 0.36, y = 0.16, z = 0.48 (a 20 % ee ligand paired at random) and run to t = 50. The final x − y over x + y is the product ee with g = 0 and ee_max = 100 %: about 0.80.

What you can use it for

  • Mechanism studies: paste your measured ligand ee and product ee pairs into the chiral amplification lab and see whether an ML2 picture fits, and with what K and g.
  • Process chemistry: estimate whether a cheaper, less pure ligand would still give acceptable product ee, and at what cost in rate.
  • Teaching equilibrium and kinetics: the exchange RR + SS ⇌ 2 RS is a neat, fully solvable example for the ODE solver.
  • Next: what happens when the product is its own catalyst, and a tiny excess snowballs into almost one hand, in part 3.

Sources

  • Nobel Prize in Chemistry 2026, announced 7 October 2026: Henri B. Kagan and Kenso Soai, 'for the discovery of non-linear effects and autocatalysis in asymmetric organic synthesis'. NobelPrize.org and the Royal Swedish Academy of Sciences (kva.se); ASBMB Today (2026).
  • Puchot C, Samuel O, Duñach E, Zhao S, Agami C, Kagan HB (1986), 'Nonlinear effects in asymmetric synthesis. Examples in asymmetric oxidations and aldolization reactions', J. Am. Chem. Soc. 108: 2353-2357.
  • Guillaneux D, Zhao S-H, Samuel O, Rainford D, Kagan HB (1994), 'Nonlinear effects in asymmetric catalysis', J. Am. Chem. Soc. 116: 9430-9439 (the ML2 and MLn models).
  • Kitamura M, Okada S, Suga S, Noyori R (1989), 'Enantioselective addition of dialkylzincs to aldehydes promoted by chiral amino alcohols', J. Am. Chem. Soc. 111: 4028-4036.
  • Wynberg H, Feringa B (1976), 'Enantiomeric recognition and interactions', Tetrahedron 32: 2831-2834.
  • Satyanarayana T, Abraham S, Kagan HB (2009), 'Nonlinear effects in asymmetric catalysis', Angew. Chem. Int. Ed. 48: 456-494 (review).
  • Blackmond DG (2000), 'Kinetic aspects of nonlinear effects in asymmetric catalysis', Acc. Chem. Res. 33: 402-411.
  • Hoang L, Bahmanyar S, Houk KN, List B (2003), J. Am. Chem. Soc. 125: 16-17 (proline-catalysed aldol re-examined: no non-linear effect).

If you are an author or publisher and want a credit or a passage changed or removed, tell us and we will act quickly.

Tools in this article

Read next

  • Channelrhodopsin: the protein that is both a light sensor and a gate Nobel series ·

    How Georg Nagel, Peter Hegemann and Ernst Bamberg put two algal genes into frog eggs and human kidney cells and found a channel opened by light: channelrhodopsin-1 in 2002 and channelrhodopsin-2 in 2003. The evidence, the physics of the photon and the pore, rebuilt with fibub tools.

  • How IceCube was built and found neutrinos from deep space Nobel series ·

    5,160 light sensors frozen into the South Pole ice, a flood of background a million times larger than the signal, and the 2013 discovery of neutrinos from beyond the solar system. Play with an event display and check the statistics yourself.

  • Mirror-image molecules and the race to make just one Nobel series ·

    Why molecules come in left- and right-handed forms, why life and medicines care which one you have, and how chemists learned to make just one. The background to the 2026 Chemistry Nobel, with animations and checks you can run.

  • The Soai reaction: how a tiny excess becomes one hand, and why life is homochiral Nobel series ·

    Kenso Soai found a reaction whose product copies its own handedness, turning a millionth of a percent excess into almost pure single-handed molecules. How it works, the triggers that tip it, and what it says about the origin of life's one-handedness.

  • Calibrating a gamma-ray detector with natural rock standards Discovery ·

    A 2025 study calibrates a germanium detector with cheap reference powders of potassium, uranium and thorium instead of a costly mixed source. We refit its efficiency curve and check the source's decay with fibub's tools.