Skip to content
LumenaryAcademy
Frontier 85 min

New reactions and framework materials

Click and bioorthogonal chemistry, light and electricity as reagents, metal–organic and covalent organic frameworks, single-atom catalysts and skeletal editing — what each can do in 2026, and how we know.

Code tested · 6AI-draftedAwaiting review by a chemist

Your guides: Chloe · The Chemist, Margaret · The Analyst, Gideon · The Sceptic

MotivationH

Two prizes and a toolkit

Two Nobel Prizes bracket this lesson. The 2022 prize went to Carolyn Bertozzi, Morten Meldal and K. Barry Sharpless for click chemistry and bioorthogonal chemistry: ways of joining molecules so reliably that they work in water, on proteins, even inside a living animal. The 2025 prize went to Susumu Kitagawa, Richard Robson and Omar Yaghi for metal–organic frameworks: crystals of metal clusters and organic struts with more internal surface per gram than any natural material.

Neither prize was for a new element or a new law. Both were for new ways of building: a small set of dependable connections, and the discipline to design with them. Over the same decades chemists added reagents that were not on the shelf before — visible photons, and electrons delivered by an electrode — learned to place a catalyst's metal atom by atom, and began to edit a molecule's skeleton rather than only its periphery.

This lesson is a briefing on those five developments. For each, the questions are the ones a colleague would ask: what is the idea, what numbers govern it, what has been demonstrated, and what is still hope. Reactions are described at the level of ideas and strategy; none of this is a bench procedure, and all of it belongs in properly equipped laboratories.

DefinitionsH

Click chemistry: a few good reactions

In 2001 Sharpless, Hartmuth Kolb and M. G. Finn argued that synthesis was ambitious in the wrong way. Nature builds its variety from a few reliable linkages (amides, glycosides, phosphodiesters); chemists prized difficult reactions and long routes. Why not restrict yourself to a handful of near-perfect reactions and search for function among what they can make?

Definition— Click reaction

A click reaction joins two building blocks quickly and essentially completely under simple conditions. In the criteria Kolb, Finn and Sharpless proposed, it is modular and wide in scope; gives very high yields; tolerates water and air; makes only harmless by-products, so that products can be isolated without chromatography; and is "spring-loaded", with a large thermodynamic driving force (they suggested more than about 84 kJ/mol), so that it goes to completion and does not reverse.

The reaction that became the emblem was found independently in 2002 by Meldal's group in Copenhagen and by Sharpless and Valery Fokin's group at Scripps. Organic azides (R–N₃) and terminal alkynes (R′–C≡C–H) had long been known to combine in a 1,3-dipolar cycloaddition, studied by Rolf Huisgen in the 1960s, to form a five-membered 1,2,3-triazole ring. Without a catalyst the reaction needs prolonged heating and gives a mixture of two regioisomers. With copper(I) it runs at room temperature, often in water, roughly seven orders of magnitude faster, and gives only the 1,4-disubstituted triazole: the copper-catalysed azide–alkyne cycloaddition, CuAAC.

R–N₃+H–C≡C–R′an azidea terminal alkyneCu(I) catalystwater, room temp.1,4-disubstituted 1,2,3-triazoleNNNRR′H12345
The copper-catalysed azide–alkyne cycloaddition (CuAAC). An organic azide and a terminal alkyne give a single product, a 1,4-disubstituted 1H-1,2,3-triazole: the azide's substituent R ends up on N1 and the alkyne's R′ on C4. Without copper the same partners need heat and give a mixture of the 1,4- and 1,5-isomers.

The cycle runs through a copper acetylide, and isotope-labelling experiments in 2013 showed two copper atoms cooperating in the ring-forming step; ruthenium catalysts give the complementary 1,5-isomer. The triazole is a good product: aromatic, robust to hydrolysis, oxidation and reduction, and similar in size and dipole to an amide bond, which medicinal chemists exploit.

CuAAC now attaches dyes, polymers and drugs to proteins, functionalises surfaces, builds screening libraries, and labels newly made DNA in dividing cells. In 2014 Sharpless's group added sulfur(VI) fluoride exchange (SuFEx), in which unusually stable S(VI)–F bonds react cleanly only when specifically activated.

Note

Low-molecular-weight organic azides store a great deal of energy, and some are shock- or heat-sensitive. Industrial and academic laboratories that use click chemistry follow written hazard assessments that limit which azides are made, in what quantity and with what shielding. This is part of why click chemistry is a laboratory method, not a classroom one.

Worked examplesH

Bioorthogonal chemistry: a race for rate constants

A living cell is a crowded solution of thiols, amines, carboxylates, metal ions and water at 37 °C. Carolyn Bertozzi asked whether one could run a reaction inside that crowd between two partners that ignore everything else, and called such reactions bioorthogonal. The classic experiment is metabolic labelling: feed cells a sugar carrying a small azide, let their enzymes build it into cell-surface glycans, then add a probe that reacts only with azides.

The partners must be inert to biology, non-toxic, and fast at low concentration, and these pull against each other. Bertozzi's modified Staudinger ligation of an azide with a phosphine (2000) met the first two and failed the third. CuAAC is fast, but copper(I) is toxic to cells. The answer in 2004 was the strain-promoted azide–alkyne cycloaddition (SPAAC): bend the alkyne into an eight-membered ring and the stored strain drives the reaction without copper. In 2008 Joseph Fox's and Ralph Weissleder's groups introduced a much faster family, the inverse-electron-demand Diels–Alder reaction of 1,2,4,5-tetrazines with strained alkenes such as trans-cyclooctene (TCO), which releases N₂ and cannot reverse.

Every one of these is a second-order reaction, rate =k[A][B]= k[\text{A}][\text{B}], and in biology the concentrations are micromolar or less. With the probe B in excess, the target A decays with a pseudo-first-order half-life t1/2=ln⁡2/(k[B])t_{1/2} = \ln 2 / (k[\text{B}]). The program compares representative rate constants (orders of magnitude only; each family spans a range) and checks the shortcut against the exact second-order kinetics.

python
import numpy as np

# Representative second-order rate constants in M^-1 s^-1. These are orders of magnitude:
# the exact value depends on the particular reagents, the solvent and the temperature.
reactions = {
    "Staudinger ligation (2000)":           2e-3,
    "first cyclooctyne, SPAAC (2004)":      2e-3,
    "later cyclooctynes, SPAAC":            0.3,
    "CuAAC with a copper-binding ligand":   50.0,
    "tetrazine + trans-cyclooctene":        2e4,
}

def half_life(k, c_excess):
    """Half-life (s) of the scarce partner when the other is in excess at c_excess (mol/L)."""
    return np.log(2) / (k * c_excess)

def readable(t):
    for unit, s in (("years", 3.156e7), ("days", 86400), ("h", 3600), ("min", 60)):
        if t >= s:
            return f"{t / s:6.1f} {unit:5s}"
    return f"{t:6.1f} s    "

print(f"{'reaction':36s} {'k':>7s}   t1/2, probe at 100 µM   t1/2, probe at 1 µM")
for name, k in reactions.items():
    print(f"{name:36s} {k:7.0e}   {readable(half_life(k, 100e-6)):>21s}   {readable(half_life(k, 1e-6)):>19s}")

# Check the pseudo-first-order shortcut against the exact second-order kinetics for a
# target at 1 µM and a probe at 10 µM, integrating d[A]/dt = -k[A][B] with RK4.
k, A0, B0 = 0.3, 1e-6, 10e-6
def rate(a):
    return -k * a * (B0 - (A0 - a))          # [B] falls by as much as [A] does
a, dt, t = A0, 10.0, 0.0
while a > A0 / 2:
    k1 = rate(a); k2 = rate(a + dt * k1 / 2); k3 = rate(a + dt * k2 / 2); k4 = rate(a + dt * k3)
    a += dt * (k1 + 2 * k2 + 2 * k3 + k4) / 6
    t += dt
exact = np.log((2 * B0 - A0) / B0) / (k * (B0 - A0))   # closed form for half-conversion of A
print(f"\nSPAAC, 1 µM target + 10 µM probe, time to label half the target:")
print(f"  RK4 {t / 3600:.2f} h   exact {exact / 3600:.2f} h   pseudo-first-order {half_life(k, B0) / 3600:.2f} h")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

The table is the field's history in seven orders of magnitude. At 1 µM a Staudinger ligation needs over a decade to reach half-conversion, a good cyclooctyne nearly a month, a tetrazine–TCO pair half a minute. That is why tetrazines dominate experiments in live animals, where probes are dilute and cleared within hours.

The newest use turns ligation around. In click-to-release chemistry a drug is masked as a carbamate on a trans-cyclooctene, and reaction with a tetrazine frees it (Versteegen et al., 2013), so a toxic drug can be activated only where a tetrazine has been placed, such as a tumour. By its developers' account, this became the first bioorthogonal reaction tested in people, in an early-phase cancer trial by the US company Shasqi from 2020; pretargeted imaging and radiotherapy with tetrazines have also reached early trials. As of late 2026 none is an approved medicine.

DefinitionsH

Light and electricity as reagents

Most reactions in Reaction mechanisms move electrons in pairs. Many useful bonds are easier to make one electron at a time, through radicals, but classical radical chemistry relied on toxic tin hydrides, peroxides or ultraviolet lamps. Two methods now make single-electron chemistry routine under mild conditions.

Photoredox catalysis

A photoredox catalyst absorbs visible light and, in its excited state, transfers an electron to or from a substrate. The classic is tris(2,2′-bipyridyl)ruthenium(II), Ru(bpy)₃²⁺, whose charge-transfer excited state lives about a microsecond, long enough to meet a substrate. Studied since the 1970s for solar energy, it became a general synthetic tool through papers from David MacMillan's and Tehshik Yoon's groups in 2008 and Corey Stephenson's in 2009. Iridium complexes and organic dyes such as 4CzIPN, a carbazolyl dicyanobenzene, followed.

The key number is how strongly the excited catalyst oxidises or reduces.

Theorem— Excited-state redox potentials

Let E(PC+/PC)E(\text{PC}^{+}/\text{PC}) and E(PC/PC−)E(\text{PC}/\text{PC}^{-}) be the ground-state potentials of a catalyst PC, and let E0,0E_{0,0} be the energy of its lowest excited state PC* in electronvolts. Neglecting small entropic and solvation differences,

E(PC+/PC∗)=E(PC+/PC)−E0,0,E(PC∗/PC−)=E(PC/PC−)+E0,0.E(\text{PC}^{+}/\text{PC}^{*}) = E(\text{PC}^{+}/\text{PC}) - E_{0,0}, \qquad E(\text{PC}^{*}/\text{PC}^{-}) = E(\text{PC}/\text{PC}^{-}) + E_{0,0}.

Proof

Removing an electron from PC* is removing it from PC after first supplying the excitation energy, so the free energy of PC∗→PC++e−\text{PC}^{*} \to \text{PC}^{+} + e^{-} is that of PC→PC++e−\text{PC} \to \text{PC}^{+} + e^{-} minus E0,0E_{0,0} (per molecule, in eV). Since a reduction potential is minus that free energy per electron charge, the oxidation couple moves down by E0,0E_{0,0} volts: the excited state is a better reductant. Adding an electron to PC* releases the excitation energy as well, so the reduction couple moves up by E0,0E_{0,0}: the excited state is a better oxidant. This is a Hess's-law cycle, the same bookkeeping as in Thermochemistry.

∎

+10−1−2E / V vs SCE (more oxidising at the top)Ru(bpy)₃²⁺Ir(ppy)₃+1.29 PC⁺/PC+0.77 PC*/PC⁻−0.81 PC⁺/PC*−1.33 PC/PC⁻+0.77 PC⁺/PC+0.31 PC*/PC⁻−1.73 PC⁺/PC*−2.19 PC/PC⁻2.10 eV2.50 eV
Potential ladder for two photoredox catalysts (V against the saturated calomel electrode, acetonitrile). Solid lines are ground-state couples; dashed lines are couples of the excited state, each shifted by the excitation energy E₀,₀. Light makes Ru(bpy)₃²⁺ both a better oxidant (+0.77 V) and a better reductant (−0.81 V); Ir(ppy)₃ trades oxidising power for a strongly reducing excited state (−1.73 V).

Combined with the electron-transfer free energy ΔGET=−F [E(acceptor)−E(donor)]\Delta G_{\text{ET}} = -F\,[E(\text{acceptor}) - E(\text{donor})], this lets a chemist choose a catalyst before running a reaction.

Electrosynthesis

An electrode is a reagent whose strength you set with a dial. The potential determines which electron transfers are possible (Electrochemistry), and Faraday's law fixes the charge, Q=nzF/εFQ = nzF/\varepsilon_F for nn moles of product needing zz electrons each at Faradaic efficiency εF\varepsilon_F. No stoichiometric oxidant or reductant becomes waste. The idea is old — Kolbe coupled carboxylates at an anode in 1849, and Manuel Baizer's reductive coupling of acrylonitrile to hexanedinitrile (adiponitrile, a nylon precursor), run by Monsanto from the 1960s, is still the largest organic electrosynthesis — but laboratory use took off only after about 2015, with standardised cells, better electrodes and combinations with nickel catalysis (Yan, Kawamata and Baran, 2017).

python
F = 96485.332          # C/mol, the Faraday constant
HC = 1239.84           # h*c in eV nm: photon energy (eV) = 1239.84 / wavelength (nm)

# Ground-state potentials E(PC+/PC) and E(PC/PC-) in V vs SCE (acetonitrile), and the excited-state
# energy E00 in eV. Approximate values as tabulated in reviews; they shift by tens of mV with conditions.
catalysts = {
    "Ru(bpy)3 2+":   (+1.29, -1.33, 2.10),
    "fac-Ir(ppy)3":  (+0.77, -2.19, 2.50),
    "4CzIPN":        (+1.52, -1.21, 2.56),
}

print(f"{'catalyst':14s} {'E00/eV':>6s} {'λ/nm':>5s}  {'E(PC+/PC*)':>10s}  {'E(PC*/PC-)':>10s}")
excited = {}
for name, (e_ox, e_red, e00) in catalysts.items():
    e_star_ox = e_ox - e00      # the excited state gives up an electron this easily: a strong reductant
    e_star_red = e_red + e00    # ... and accepts one this readily: a strong oxidant
    excited[name] = (e_star_ox, e_star_red)
    print(f"{name:14s} {e00:6.2f} {HC / e00:5.0f}  {e_star_ox:+10.2f}  {e_star_red:+10.2f}")

# Electron transfer from a donor D to an acceptor A: dG = -F [E(A/A-) - E(D+/D)] (work terms ignored).
def dG_kJ(E_acceptor, E_donor):
    return -F * (E_acceptor - E_donor) / 1000

donor = 0.90       # an illustrative electron donor, E(D+/D) = +0.90 V vs SCE
acceptor = -1.50   # an illustrative electron acceptor, E(A/A-) = -1.50 V vs SCE
print("\nΔG (kJ/mol) for the excited catalyst to ...")
print(f"{'':14s} {'oxidise D (+0.90 V)':>20s} {'reduce A (-1.50 V)':>20s}")
for name, (e_star_ox, e_star_red) in excited.items():
    print(f"{name:14s} {dG_kJ(e_star_red, donor):20.0f} {dG_kJ(acceptor, e_star_ox):20.0f}")

# Faraday's law for an electrosynthesis: charge, time and electrical energy per tonne of product.
M = 108.14          # g/mol, hexanedinitrile (adiponitrile), made by a 2-electron reductive coupling
z, FE = 2, 0.90     # electrons per molecule; assumed Faradaic efficiency
V_cell = 4.0        # V, an assumed cell voltage
n = 1e6 / M                         # mol in one tonne
Q = n * z * F / FE                  # C
print(f"\nper tonne of product: {n:.0f} mol, charge {Q:.3e} C, "
      f"electrical energy {Q * V_cell / 3.6e9:.2f} MWh at {V_cell} V")
print(f"one tonne a day needs a total current of {Q / 86400 / 1e3:.1f} kA")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

For the illustrative donor at +0.90 V, excited Ru(bpy)₃²⁺ is 13 kJ/mol short, while 4CzIPN oxidises it with 43 kJ/mol to spare; to reduce the acceptor at −1.50 V, only Ir(ppy)₃ will do. And industrial electrochemistry is an exercise in current: a tonne a day of a two-electron product needs about 23 kA.

Metallaphotoredox couplings (photoredox plus nickel, from 2014) are now routine in pharmaceutical discovery. Scale-up is the obstacle for both methods: light penetrates only millimetres into a coloured solution (Beer–Lambert), and current flows only at electrode surfaces, so production means thin flow channels and large electrode areas rather than bigger flasks.

DefinitionsH

Reticular chemistry: metal–organic and covalent organic frameworks

Definition— Metal–organic framework (MOF)

A metal–organic framework is a crystalline solid in which metal ions or metal-oxo clusters (the nodes, or secondary building units) are joined by organic linkers through coordination bonds into an extended network with permanent, accessible pores. A covalent organic framework (COF) is the analogous crystal built from light elements (B, C, N, O) joined entirely by covalent bonds. Designing such crystals by choosing nodes, linkers and the underlying net is called reticular chemistry.

node: a metal cluster(secondary building unit)linker: a rigid organicmolecule bonded at both endspore: about 1 to 3 nm,open to gases and solvents
A metal–organic framework in two dimensions. Metal clusters (squares) sit at the nodes of a net and rigid organic linkers (hexagons stand for benzene rings) join them, leaving regular pores (shaded) between. In MOF-5 the same idea runs in three dimensions: Zn₄O clusters at the corners of cubes, joined by benzene-1,4-dicarboxylate linkers.

From coordination polymers to designed crystals

In 1989 Bernard Hoskins and Richard Robson in Melbourne linked copper(I) ions with a tetrahedral tetranitrile into a diamond-like network and argued that crystals with large cavities could be built by design. Early frameworks tended to collapse when their solvent was removed. In 1997 Susumu Kitagawa's group in Kyoto showed reversible gas uptake in a coordination polymer, and he went on to describe flexible frameworks that "breathe" as molecules enter. In 1999 Omar Yaghi's group reported MOF-5, Zn₄O(BDC)₃ (BDC = benzene-1,4-dicarboxylate), which stayed crystalline and porous without its solvent; in 2002 a whole "isoreticular" family on the same cubic net with linkers of different lengths; and in 2005 the first COFs, crystallised through reversible covalent bonds that let mistakes heal.

Reticular chemistry thinks in nets, the graphs underlying crystal structures: MOF-5 is the primitive cubic net of six-connected nodes; HKUST-1, from copper paddlewheels and benzene-1,3,5-tricarboxylate, is a net of three- and four-connected nodes. The node geometry and linker angle largely choose the net, and the linker length sets the pore size: graph theory with bond lengths. Tens of thousands of MOFs are known, from groups on every continent: UiO-66 (Oslo), MIL-101 (Versailles), the zeolitic imidazolate frameworks (Los Angeles), the DUT series (Dresden), and many from China, Japan, Korea and India.

How much surface?

The program computes a benchmark (a graphene sheet, counted on both faces), the crystal density of MOF-5 from its unit cell, and the BET area every MOF paper reports: fit the Brunauer–Emmett–Teller equation to a nitrogen isotherm at 77 K and multiply the monolayer amount by the area of one N₂ molecule, 0.162 nm².

python
import numpy as np

NA = 6.02214076e23
SIGMA_N2 = 0.162e-18     # m^2 occupied by one adsorbed N2 molecule at 77 K (the conventional value)

# 1. The graphene benchmark: a sheet of carbon, both faces counted.
a = 2.46e-10                                   # m, graphene lattice constant
cell = np.sqrt(3) / 2 * a**2                   # m^2 per unit cell, which holds 2 C atoms
graphene = 2 * cell / 2 * NA / 12.011          # two faces, per C atom, per gram
print(f"graphene, both faces: {graphene:.0f} m^2/g")

# 2. Crystal density of MOF-5, Zn4O(BDC)3: cubic, a = 25.83 Å, 8 formula units per cell.
M_mof5 = 4 * 65.38 + 16.00 + 3 * (8 * 12.011 + 4 * 1.008 + 4 * 15.999)
rho = 8 * M_mof5 / NA / (25.83e-8) ** 3        # g/cm^3
print(f"MOF-5: M = {M_mof5:.1f} g/mol, crystal density {rho:.3f} g/cm^3;"
      f" 3800 m^2/g is {3800 * rho:.0f} m^2 per cm^3 of crystal")

# 3. BET analysis. The linearised BET equation, with x = p/p0 and n the amount adsorbed:
#    x / (n (1 - x)) = 1/(n_m C) + (C - 1)/(n_m C) * x
def bet_fit(x, n):
    slope, intercept = np.polyfit(x, x / (n * (1 - x)), 1)
    return 1 / (slope + intercept), 1 + slope / intercept     # n_m, C

def area(n_m_mol_per_g):
    return n_m_mol_per_g * NA * SIGMA_N2

# Check the fit on an isotherm that obeys the BET equation exactly (n_m = 10 mmol/g, C = 100).
x = np.linspace(0.05, 0.30, 11)
n_bet = 10e-3 * 100 * x / ((1 - x) * (1 - x + 100 * x))
n_m, C = bet_fit(x, n_bet)
print(f"\nexact BET isotherm: n_m = {n_m * 1e3:.2f} mmol/g, C = {C:.1f}, area {area(n_m):.0f} m^2/g")

# A microporous framework fills its pores at low pressure (a Type I isotherm). Model it as
# strongly binding sites, a second step near x = 0.05, and a little external surface.
xs = np.geomspace(1e-4, 0.4, 60)
n_mof = (30e-3 * 800 * xs / (1 + 800 * xs) + 14e-3 * (xs / 0.05) ** 4 / (1 + (xs / 0.05) ** 4)
         + 1e-3 * xs / (1 - xs))
top = int(np.argmax(n_mof * (1 - xs)))   # Rouquerol: use only points where n(1 - x) still rises
for lo, hi in [(0.05, 0.30), (0.005, 0.05), (0.01, xs[top])]:
    m = (xs >= lo) & (xs <= hi)
    n_m, C = bet_fit(xs[m], n_mof[m])
    print(f"BET fitted on x = {lo:5.3f}-{hi:5.3f}: C = {C:6.0f}, 'area' {area(n_m):5.0f} m^2/g")
print(f"n(1-x) peaks at x = {xs[top]:.3f}")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

Graphene's 2,628 m²/g is not a ceiling: by exposing edges a stacked sheet would hide, frameworks have passed it, with BET areas around 7,000 m²/g for NU-110 (2012) and 7,800 m²/g for DUT-60 (2018). MOF-5's density of 0.59 g/cm³ makes volumetric area (2,255 m² per cm³) the fairer figure for a tank of fixed size. And the same simulated isotherm gives 3,183, 3,347 or 4,080 m²/g depending on the pressure range fitted; the textbook range of 0.05 to 0.30 even returns a meaningless negative CC.

ComputationC

Storing gas in a framework

Adsorbed molecules pack more densely than gas at the same pressure, so a tank filled with a porous solid can hold more. How strongly should the solid bind? Too weakly and little is adsorbed; too strongly and the gas will not come out when the tank is drained to the lowest usable pressure.

Theorem— Optimal Langmuir affinity for storage

Suppose adsorption follows a Langmuir isotherm q(p)=qmax⁡ bp/(1+bp)q(p) = q_{\max}\, bp/(1+bp), and the tank is filled to pHp_H and emptied to pLp_L. The deliverable amount D(b)=q(pH)−q(pL)D(b) = q(p_H) - q(p_L) is largest when b=1/pHpLb = 1/\sqrt{p_H p_L}, and then D=qmax⁡ (r−1)/(r+1)D = q_{\max}\,(r-1)/(r+1) with r=pH/pLr = \sqrt{p_H/p_L}.

Proof

Differentiate: D′(b)=qmax⁡[pH(1+bpH)2−pL(1+bpL)2]D'(b) = q_{\max}\left[\dfrac{p_H}{(1+bp_H)^2} - \dfrac{p_L}{(1+bp_L)^2}\right]. Setting it to zero and taking positive square roots gives pH (1+bpL)=pL (1+bpH)\sqrt{p_H}\,(1 + bp_L) = \sqrt{p_L}\,(1 + bp_H), so pH−pL=bpHpL (pH−pL)\sqrt{p_H} - \sqrt{p_L} = b\sqrt{p_H p_L}\,(\sqrt{p_H} - \sqrt{p_L}) and b=1/pHpLb = 1/\sqrt{p_H p_L}. DD is zero at b=0b = 0 and tends to zero as b→∞b \to \infty, so this single stationary point is the maximum. Substituting, bpH=rbp_H = r and bpL=1/rbp_L = 1/r, so D/qmax⁡=r/(1+r)−1/(1+r)=(r−1)/(r+1)D/q_{\max} = r/(1+r) - 1/(1+r) = (r-1)/(r+1).

∎

python
import numpy as np

R, T = 8.314, 298.15
B_CH4 = -43e-6            # m^3/mol, approximate second virial coefficient of methane at 298 K
V_STP = 22.414            # cm^3 of gas at 0 °C and 1 atm per mmol

def gas_mmol_per_cm3(p_bar):
    """Methane in an empty tank: Z = 1 + Bp/RT (good to a few per cent below about 100 bar)."""
    p = p_bar * 1e5
    Z = 1 + B_CH4 * p / (R * T)
    return p / (Z * R * T) / 1e3          # mol/m^3 -> mmol/cm^3

def langmuir(p_bar, q_max, b):
    """Amount adsorbed, mmol/g, for a Langmuir isotherm with affinity b (1/bar)."""
    return q_max * b * p_bar / (1 + b * p_bar)

p_high, p_low = 65.0, 5.8      # bar: fill pressure, and the pressure below which an engine cannot draw gas
q_max, rho_crystal, void = 12.0, 0.88, 0.60   # an illustrative framework: mmol/g, g/cm^3, pore fraction

def deliverable_vv(b):
    """Gas delivered between p_high and p_low per volume of tank packed with crystals, in cm^3(STP)/cm^3."""
    adsorbed = (langmuir(p_high, q_max, b) - langmuir(p_low, q_max, b)) * rho_crystal
    in_pores = void * (gas_mmol_per_cm3(p_high) - gas_mmol_per_cm3(p_low))
    return (adsorbed + in_pores) * V_STP

empty = (gas_mmol_per_cm3(p_high) - gas_mmol_per_cm3(p_low)) * V_STP
print(f"empty tank, 65 -> 5.8 bar: Z(65 bar) = {1 + B_CH4 * 65e5 / (R * T):.3f}, delivers {empty:.0f} v/v")
for b in [0.005, 0.02, 0.05, 0.2, 1.0]:
    print(f"framework with b = {b:5.3f} /bar: delivers {deliverable_vv(b):5.0f} v/v"
          f"  (stays behind at 5.8 bar: {langmuir(p_low, q_max, b) * rho_crystal * V_STP:4.0f} v/v)")

bs = np.geomspace(1e-3, 10, 4001)
best = bs[np.argmax([deliverable_vv(b) for b in bs])]
print(f"\nbest b by search: {best:.4f} /bar;  1/sqrt(p_high p_low) = {1 / np.sqrt(p_high * p_low):.4f} /bar")
r = np.sqrt(p_high / p_low)
print(f"at the optimum the adsorbed part delivers (r-1)/(r+1) = {(r - 1) / (r + 1):.3f} of q_max")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

With these illustrative parameters a packed tank delivers 164 volumes of methane per volume of tank, against 61 empty. Stronger binding is not better: at b=1b = 1 bar⁻¹ the framework keeps 202 v/v it never gives back. The search lands on 1/pHpL=0.05151/\sqrt{p_H p_L} = 0.0515 bar⁻¹, as the theorem says.

Translated into energy, the theorem explains a decade of hydrogen-storage results. For the Langmuir constant to sit at its optimum near room temperature, Bhatia and Myers estimated that hydrogen needs an adsorption enthalpy of about 15 kJ/mol, while physisorption of H₂ on most frameworks gives only 4 to 8 kJ/mol. That is why MOF hydrogen storage works well at 77 K and poorly at 298 K, and why research targets open metal sites that bind H₂ more strongly.

Real tanks lose much of this on-paper advantage: powders pack at perhaps half the crystal density, and the heat released on filling cuts uptake. The applications on the market as of 2026 are ones where such penalties matter less: cylinders that hold hazardous dopant gases for chip manufacturing below atmospheric pressure, so that a leak draws air in rather than pushing gas out; and carbon-capture filters built on CALF-20, a zinc triazolate–oxalate framework from the University of Calgary that BASF announced in 2023 it could make at hundreds of tonnes a year. Water harvesting from desert air and capturing CO₂ from open air with amine-bearing COFs (COF-999, 2024) are being scaled; Sustainable chemistry asks what such capture costs in energy.

DefinitionsH

Single-atom catalysts

A supported metal catalyst works only at its surface. For ideal cuboctahedral clusters with mm atoms along each edge, the total number of atoms is (10m3−15m2+11m−3)/3(10m^3 - 15m^2 + 11m - 3)/3 and the number on the surface is 10m2−20m+1210m^2 - 20m + 12: 12 of 13 atoms for m=2m = 2, 42 of 55 (76 %) for m=3m = 3, 92 of 147 (63 %) for m=4m = 4, and the fraction keeps falling as the particle grows. For platinum, iridium or palladium, every buried atom is expensive metal doing nothing.

Definition— Single-atom catalyst

A single-atom catalyst contains its active metal only as isolated atoms (usually ions) anchored to a support — an oxide, nitrogen-doped carbon, or the surface of another metal — with no metal–metal bonds between them. Every metal atom is in principle accessible, and its electronic structure is set by the atoms that hold it, much as a ligand set tunes a metal complex in Coordination chemistry.

The term was coined by Tao Zhang, Jun Li, Jingyue Liu and colleagues in 2011 for platinum atoms on iron oxide that oxidised CO (Qiao et al.). Relatives include single-atom alloys, a few palladium or platinum atoms in a copper surface that split H₂ while the copper does the rest (Kyriakou et al., 2012), and M–N–C materials, iron or cobalt held by four nitrogens in a carbon sheet, studied as platinum-free fuel-cell cathodes. The clearest industrial success predates the name: Graham Hutchings's gold-on-carbon catalyst for making chloroethene (vinyl chloride) from acetylene, commercialised with Johnson Matthey in China in the mid-2010s to replace mercury(II) chloride, works through atomically dispersed gold cations.

"Single atom" is a claim about every atom in the sample, and no technique sees every atom. Electron microscopy shows isolated dots, but in fields a few tens of nanometres wide; X-ray absorption (EXAFS) averages over the sample and should show no metal–metal neighbours; adsorbed CO should bind only linearly, never bridging two atoms. A convincing paper uses all three, and again after reaction, because isolated atoms often sinter into clusters when hot.

The honest limits in 2026: loadings are usually a few per cent by mass or less, because isolated atoms are stable only at strong anchoring sites; stability under industrial conditions is the central problem ("atom trapping" on ceria is one answer); and reactions needing adjacent metal atoms cannot run on a single atom. Single atoms are a tool for selectivity and metal economy, not a universal upgrade.

Worked examplesH

Skeletal editing

Most of synthesis edits a molecule's periphery: replace a C–H with a C–F, add a methyl. The skeleton — which atoms make up the rings and chains — is fixed early in a route. Yet a medicinal chemist often wants exactly a skeletal change: one ring carbon replaced by nitrogen, which shifts basicity, solubility and metabolism (Medicinal chemistry), or a six-membered ring contracted to five. Until recently each such analogue meant a new synthesis (Synthesis and retrosynthesis).

Skeletal editing is the family of reactions that insert, delete or exchange single atoms of a molecule's core. The Ciamician–Dennstedt rearrangement of 1881 expanded pyrroles into pyridines, in poor yield; the modern field took shape around 2021. Mark Levin's group in Chicago reported deleting the nitrogen of secondary amines, joining the two carbons that flanked it, and a general carbon-atom insertion into pyrroles and indoles; Bill Morandi's group in Zürich inserted nitrogen into indoles to give quinazolines and quinoxalines; and groups in Germany, China, Japan and elsewhere have reported swaps that turn particular pyridines into benzenes, and the reverse.

NHNHNNdelete Nring contractsinsert Cpyrrole to pyridineswap N for Cpyridine to benzene
Three kinds of single-atom skeletal edit, shown on simple rings. Deleting the nitrogen of a cyclic secondary amine contracts the ring; inserting a carbon expands a pyrrole into a pyridine; swapping a ring nitrogen for carbon turns a pyridine into a benzene ring. Each is known for particular substrates, not as a general reaction.

The strategic value is that an edit applies to a finished, complex molecule, so one advanced intermediate yields a series of skeletal analogues; Levin, Richmond Sarpong and colleagues argue for planning with this "single-atom logic". As of 2026, each method works on particular ring systems, yields are often moderate, some reagents are energetic, and few edits are routine in industry. The area pairs naturally with the high-throughput experiments of Self-driving laboratories.

ExplanationH

The professional view

Who uses what. Pharmaceutical discovery teams use CuAAC and SuFEx for libraries and conjugates, metallaphotoredox couplings for sp³-rich analogues, electrochemistry increasingly for oxidations and couplings, and skeletal edits experimentally. Chemical biologists use tetrazine and cyclooctyne chemistry daily for imaging and proteomics. Framework materials are made commercially in Europe, North America and Asia.

Tools. Gas adsorption in porous solids is simulated with grand canonical Monte Carlo (open codes such as RASPA); pore geometry and accessible surface area are computed from crystal structures with Zeo++, by rolling an N₂-sized probe over the framework. Screening campaigns compute cheap geometry first, simulate uptake only for the survivors, and leave the hardest questions — can it be made, and does it survive water? — to chemists. Machine-learned models now pre-screen hypothetical frameworks by the hundred thousand, the theme of Machine learning for molecules and reactions and of Lumenary Materials' AI and materials discovery; for crystal structure itself, see crystal structures.

What is hype. Surface-area records do not predict usefulness; a framework that loses crystallinity in humid air will not survive a flue-gas duct. Single-atom catalysis has a large literature of activity claims on poorly characterised samples. Electrifying chemistry is real but slow, because electricity costs more per joule than heat from fuel and electrode area grows with output. Each field is a genuine advance; each is narrower than its press releases.

Common misconceptionsH

Common misconceptions

Misconception

"A click reaction is any fast, high-yielding reaction." Speed alone is not enough. Click reactions are defined by reliability across many substrates, tolerance of water and air, harmless by-products and simple isolation. Many fast reactions fail these tests; and the uncatalysed azide–alkyne cycloaddition, though highly exothermic, is too slow and unselective to qualify.

Misconception

"The higher the BET area, the better the MOF for storage or capture." Deliverable capacity depends on binding strength (optimal, not maximal), volumetric density, packing, selectivity against water and other gases, and stability over thousands of cycles. And the BET area of a microporous solid depends on how the fit was done.

Misconception

"Single-atom catalysts are always more active per atom." They use every atom, but an isolated atom has different electronic structure from a surface atom of a particle, and reactions needing ensembles of adjacent atoms may not run at all. Some single-atom catalysts are less active per atom and valuable for their selectivity instead.

Exercises

Exercises

1. New reactions and framework materials

Warm-upHuman

Bertozzi's group introduced the strain-promoted azide–alkyne cycloaddition (SPAAC) in 2004, although the copper-catalysed version (CuAAC) was already known to be fast and selective. What was the main reason?

Choose one

2. New reactions and framework materials

CoreHuman

A cell-surface glycan carries an azide. You add a cyclooctyne dye at 50 µM, far in excess, and the reaction has k=0.30k = 0.30 M⁻¹ s⁻¹. Treating it as pseudo-first-order, after how many hours is half of the azide labelled? Give the nearest whole number of hours.

3. New reactions and framework materials

CoreHuman

A new organic photocatalyst has ground-state potentials E(PC+/PC)=+1.45E(\text{PC}^{+}/\text{PC}) = +1.45 V and E(PC/PC−)=−1.60E(\text{PC}/\text{PC}^{-}) = -1.60 V (vs SCE), and an excited-state energy E0,0=2.70E_{0,0} = 2.70 eV. What is the excited-state reduction potential E(PC∗/PC−)E(\text{PC}^{*}/\text{PC}^{-}), the measure of how strongly the excited catalyst oxidises a substrate? Give it in millivolts vs SCE.

4. New reactions and framework materials

Warm-upHuman

A laboratory electrosynthesis makes 0.50 mol of a product that needs 2 electrons per molecule. The cell runs at a constant 2.0 A with a Faradaic efficiency of 80 %. How many hours does the run take, to the nearest hour? (F=96 485F = 96\,485 C/mol.)

5. New reactions and framework materials

CoreComputational

Write bet_area(x, n) that fits the linearised BET equation

xn(1−x)=1nmC+C−1nmC x\frac{x}{n(1-x)} = \frac{1}{n_m C} + \frac{C-1}{n_m C}\,x

by least squares to relative pressures x and amounts adsorbed n (mol/g), and returns a tuple (area, C): the BET area in m²/g, taking 0.162 nm² per N₂ molecule, and the BET constant CC. Accept lists or NumPy arrays.

Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

6. New reactions and framework materials

StretchComputational

A gas is stored in an adsorbent that follows the Langmuir isotherm q(p)=qmax⁡ bp/(1+bp)q(p) = q_{\max}\,bp/(1+bp), filled to p_high and emptied to p_low.

Write deliverable(q_max, b, p_high, p_low), the amount released between the two pressures, and best_b(p_high, p_low), the affinity bb that maximises it.

Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

7. New reactions and framework materials

CoreHuman

A paper claims that its platinum-on-ceria catalyst contains only isolated Pt atoms. Which observations support the claim? Select all that apply.

Choose all that apply

8. New reactions and framework materials

Warm-upHuman

Which of these transformations is a skeletal edit rather than a peripheral one?

Choose one

Further reading and resources