The Robin's Quantum Compass

Biology lens: a compass that works through the eyes

For a long time, two mechanisms competed: tiny iron-mineral particles (magnetite) in the beak, or a magnetically sensitive chemical reaction in the eye 6 7. Experiments with robins pointed to the eye. Manuela Zapka and colleagues found that robins with bilateral lesions of Cluster N, a night-active, light-processing region of the forebrain, could no longer orient with their magnetic compass – but they could still use their sun and star compasses. Cutting the ophthalmic branch of the trigeminal nerve, which connects the beak to the brain, had no effect on the magnetic compass 7. According to the authors, the data strongly suggest that the robin’s magnetic compass is mediated by vision 7.

According to the leading hypothesis, the sensor molecule is cryptochrome. Jingjing Xu and colleagues produced cryptochrome 4 of the European robin and showed that its photochemistry is magnetically sensitive in the test tube – more so than cryptochrome 4 from two birds that do not migrate, the chicken and the pigeon. Mutations revealed that four successive radical pairs inside the protein (each made of a flavin and a tryptophan) take part 1.

The robin’s world is not only magnetic but also acoustic. In cities, robins sing at night where it is noisy during the day: daytime noise predicted night-time singing much better than artificial light at night 8.

Physics lens: spins that feel a weak field

An electron behaves like a tiny magnet; its magnetic property is called spin. In Earth’s field of about 50 µT 3, the splitting between the two magnetic energy levels of an electron spin – the Zeeman splitting – is extremely small: about 2 × 10⁻⁷ of the thermal energy of a molecule at body temperature, as the computer science lens calculates. A sensor that relied on aligning spins in thermal equilibrium would give an extremely weak signal.

The radical-pair model does not need thermal equilibrium. Light creates the pair, and what matters is how the field changes the spin dynamics during the short life of the pair – the simulation in the computer science lens contains no temperature at all, and still the field changes the outcome. Theoretical work showed that fields of Earth’s strength can change the reaction yield, and that the change depends on how the radical pair is aligned with the field, if the coupling of an electron to nearby nuclear spins – the hyperfine coupling – is anisotropic 5. In the laboratory, a synthetic molecule built of a carotenoid, a porphyrin and a fullerene showed exactly this: the lifetime of its radical pair changed in fields of 50 µT or less, and the response depended on the field’s direction 3.

A test with radio waves. If the compass relies on spins, oscillating magnetic fields at the right frequencies should disturb it. Thorsten Ritz and colleagues exposed robins to radio-frequency fields: a vertically aligned broadband field (0.1–10 MHz) or a 7-MHz field disoriented the birds. With the 7-MHz field the effect depended on the angle: parallel to Earth’s field the birds oriented normally, at 24° or 48° they were disoriented – consistent with a resonance on the transitions between singlet and triplet states 6. In Earth’s field an electron spin precesses at about 1.4 MHz – inside the broadband range that disturbed the birds, though below the single frequency of 7 MHz that was tested; that the two are connected is our inference, not a result of the study.

City noise. Svenja Engels and colleagues found that robins could not use their magnetic compass in unscreened wooden huts on the campus of the University of Oldenburg. In huts screened with grounded aluminium, which weakened the electromagnetic noise between 50 kHz and 5 MHz by about two orders of magnitude, the birds oriented again – and lost it as soon as the grounding was removed or noise was added. The tests were fully double-blinded 9.

Mathematics lens: reading the tilt

Split Earth’s field vector B into a horizontal part B_h and a vertical part B_v. The inclination I is the angle between the field and the horizontal plane:

The robin’s rule uses this geometry: along the horizontal axis of the field, it takes as “poleward” the direction in which the field line and the gravity vector form the smaller angle 2. Whether the field lines point into the ground or out of it, the rule picks the side towards the nearer pole – it needs the tilt, not the polarity.

Why polarity drops out. If you reverse the field, B → −B, the field lines still lie along the same axis with the same tilt. A sensor that only responds to the axis of the field – to the angle θ between the field and the molecule, with no difference between θ and 180° − θ – cannot tell the two polarities apart. That is exactly what the robin’s compass shows 2, and it is what a radical pair predicts, as the simulation in the next lens confirms.

One consequence follows directly from the formula: where the field is horizontal (I = 0, at the magnetic equator), the field line forms the same angle with gravity on both sides – the rule alone cannot distinguish poleward from equatorward there.

Computer science lens: simulating a quantum compass

The program below simulates the simplest radical pair that can work as a compass, in the spirit of the model by Ritz, Adem and Schulten 5: two electron spins, one of them coupled to a single nucleus with an anisotropic (axial) hyperfine coupling. The pair starts as a singlet and reacts with a rate k; the program calculates the singlet yield – the fraction of pairs that react from the singlet state – for every angle θ between Earth’s field and the molecule. The field strength of 50 µT is Earth’s typical value 3; the hyperfine coupling of 1 mT, the axial shape and the lifetime of 1 µs are model assumptions. The model is deliberately minimal – one nucleus, no relaxation – and shows the principle, not the size of the effect in a real protein.

The method: the spin Hamiltonian is an 8 × 8 matrix (three spins ½). After diagonalising it, the singlet yield for an exponential lifetime is a sum over all pairs of eigenstates m, n with energy difference ω_mn: each term is weighted by k²/(k² + ω_mn²), so only spin oscillations slower than about the reaction rate matter.

import numpy as np
import matplotlib.pyplot as plt

# Physical constants
GAMMA_E = 1.760859e11     # gyromagnetic ratio of the electron (rad s⁻¹ T⁻¹)
MU_B = 9.2740101e-24      # Bohr magneton (J/T)
K_B = 1.380649e-23        # Boltzmann constant (J/K)
H = 6.62607015e-34        # Planck constant (J s)

# Model – a minimal radical pair (assumptions, see the text)
B_EARTH = 50e-6           # Earth's field (T), about 50 µT
HYPERFINE = 1.0e-3        # axial hyperfine coupling of electron 1 to one nucleus (T)
ANISOTROPY = (0.0, 0.0, 1.0)   # hyperfine tensor diag(x, y, z) in units of HYPERFINE: purely axial
LIFETIME = 1e-6           # radical-pair lifetime (s); recombination rate k = 1/LIFETIME

# Why this is surprising: the Zeeman splitting of an electron spin in Earth's field vs. heat
zeeman = 2 * MU_B * B_EARTH
print(f"Zeeman splitting / thermal energy at 37 °C: {zeeman / (K_B * 310):.0e}")
print(f"electron spins precess at {GAMMA_E * B_EARTH / (2 * np.pi) / 1e6:.1f} MHz in Earth's field")

# Spin operators for spin 1/2, combined for electron 1, electron 2 and the nucleus (8 states)
sx, sy, sz = (np.array(m, dtype=complex) / 2 for m in ([[0, 1], [1, 0]], [[0, -1j], [1j, 0]], [[1, 0], [0, -1]]))
one = np.eye(2)
def op(a, b, c):
    return np.kron(np.kron(a, b), c)
spin = (sx, sy, sz)
S1 = [op(s, one, one) for s in spin]                     # electron 1
S2 = [op(one, s, one) for s in spin]                     # electron 2
S1_I = [op(s, one, s) for s in spin]                     # products S1x·Ix, S1y·Iy, S1z·Iz
singlet = 0.25 * np.eye(8) - sum(op(s, s, one) for s in spin)     # projector onto the singlet state


def singlet_yield(theta):
    """Fraction of radical pairs that react from the singlet state, field at angle theta to the molecule."""
    b = np.array([np.sin(theta), 0.0, np.cos(theta)])
    w = GAMMA_E * B_EARTH
    zeeman_h = w * sum(bi * (s1 + s2) for bi, s1, s2 in zip(b, S1, S2))
    hyperfine_h = GAMMA_E * HYPERFINE * sum(a * si for a, si in zip(ANISOTROPY, S1_I))
    energies, states = np.linalg.eigh(zeeman_h + hyperfine_h)
    p = np.einsum("im,ij,jn->mn", states.conj(), singlet, states)   # singlet projector in the eigenbasis
    k = 1 / LIFETIME
    omega = energies[:, None] - energies[None, :]
    return float(np.sum(np.abs(p) ** 2 * k**2 / (k**2 + omega**2)).real / 2)   # Tr(singlet) = 2


angles = np.radians(np.linspace(0, 180, 181))
yields = np.array([singlet_yield(a) for a in angles])
print(f"singlet yield: {yields[0]:.4f} along the field axis, {yields[90]:.4f} across it")
print(f"compass signal (max − min): {100 * (yields.max() - yields.min()):.2f} % of all pairs")
print(f"north and south look alike – yield at θ equals yield at 180° − θ to within 1e-12: "
      f"{np.allclose(yields, yields[::-1], rtol=0, atol=1e-12)}")

plt.rcParams["font.size"] = 13      # large type: the chart stays readable on a phone
fig, ax = plt.subplots(figsize=(6.4, 5.0), layout="constrained")
ax.plot(np.degrees(angles), 100 * yields, lw=2.5)
ax.axvline(90, color="grey", ls=":", lw=1.5)
ax.set_xlabel("angle between Earth's field and the molecule (°)")
ax.set_ylabel("singlet yield (%)")
ax.set_xticks([0, 45, 90, 135, 180])
ax.set_title("Angle matters – north or south does not")
plt.show()

• Open in Colab
• Download .py
• How to run this code

Output · what the code prints, re-run at every build

Zeeman splitting / thermal energy at 37 °C: 2e-07
electron spins precess at 1.4 MHz in Earth's field
singlet yield: 0.5001 along the field axis, 0.2557 across it
compass signal (max − min): 24.44 % of all pairs
north and south look alike – yield at θ equals yield at 180° − θ to within 1e-12: True

What the result teaches:

• Tiny energy, clear signal. The Zeeman splitting is only 2 × 10⁻⁷ of the thermal energy, yet the singlet yield changes from 50 % (field along the molecule’s axis) to 25.6 % (across it): the compass signal is 24 % of all pairs in this idealised model. Spin dynamics, not equilibrium, makes this possible.
• North and south look alike. The yield at θ equals the yield at 180° − θ to within 10⁻¹² – the simulated sensor reads the axis of the field, just like the robin’s inclination compass 2.
• The right timescale. The spins precess at 1.4 MHz; a pair that lives 1 µs has time for about one turn. In our reading, this is consistent with the finding that radio-frequency fields in the megahertz range disorient robins 6.

Try it yourself:

• Make the hyperfine coupling isotropic, ANISOTROPY = (1.0, 1.0, 1.0): the compass signal disappears (0.00 %). Without a built-in direction in the molecule there is no compass.
• Shorten the lifetime to 10 ns, LIFETIME = 1e-8: the signal shrinks to 0.02 % – the pair reacts before the weak field can act.

Zeeman splitting / thermal energy at 37 °C: 2e-07
electron spins precess at 1.4 MHz in Earth's field
singlet yield: 0.3876 along the field axis, 0.3876 across it
compass signal (max − min): 0.00 % of all pairs
north and south look alike – yield at θ equals yield at 180° − θ to within 1e-12: True

Zeeman splitting / thermal energy at 37 °C: 2e-07
electron spins precess at 1.4 MHz in Earth's field
singlet yield: 0.7817 along the field axis, 0.7814 across it
compass signal (max − min): 0.02 % of all pairs
north and south look alike – yield at θ equals yield at 180° − θ to within 1e-12: True

Physical AI lens: quantum sensors and a compass made of molecules

Physical AI is about machines that sense and act in the real world. If the leading hypothesis is right, the robin’s compass would be a quantum sensor grown by evolution: a system whose quantum states respond to the quantity to be measured. In quantum sensing, a growing field of precision measurement, the working part of the sensor contains one or a few qubits, and resources such as entanglement are tailored to maximise sensitivity 10.

The chemical side is already in the laboratory. The carotenoid–porphyrin–fullerene molecule of Kiminori Maeda and colleagues showed that a synthetic radical pair can respond to fields as weak as Earth’s and to their direction – the feasibility of a chemical compass, and a guide to the design features such a sensor needs 3.

The robin also shows a lesson in design. Its compass is not a separate instrument: it may be built into vision, processed by a brain region that also handles light 7 5, and it works alongside the sun and star compasses 7. And it is fragile: radio-frequency noise from human electronics can switch it off 9. In our view, these two points – sensing built into the body, and robustness against noise – are exactly the questions a robot with a bio-inspired quantum compass would face. Such a robot does not exist yet; it is a research idea, not a product.