The two photons that are not quite back to back
Assumes: The partner that sets the price of a pair · The invariant that survives a boost
Positron emission tomography runs on the reverse of pair production. A molecule labelled with a positron-emitting isotope — most often fluorine-18 in a sugar the body takes up where it is burning fuel — is injected, and wherever it collects, its nuclei decay and throw out positrons. Each positron travels a millimetre or so through tissue, slowing down, and then meets an electron and annihilates. The partner that sets the price of a pair ended on exactly this: two photons of 511 keV leave in opposite directions, a ring of detectors catches both, and the annihilation happened somewhere on the line between them. Millions of such lines, crossing, reconstruct where the tracer went.
The phrase “opposite directions” carries an assumption, and the assumption is relativity’s. Two photons from a pair at rest must leave back to back, because their momenta must add to zero. A pair that is moving gives two photons whose momenta add to the pair’s, and they are not back to back. In tissue, the positron is at rest by the time it annihilates; the electron is not. The question is how much the electron’s motion matters, and the answer is a number that sets the sharpness of every PET image and also, in a different laboratory, maps the inside of a metal.
Where the 511 comes from
Each photon’s energy is the electron’s rest energy, 511 keV, because the pair has almost no kinetic energy to add: the whole of what is converted is the two masses, as mass is a form of energy put it, and the photons carry it away in two equal shares. That equality is itself a statement about the frame. Two photons that share a pair’s energy equally and leave in opposite directions are what a pair at rest must produce, because only equal energies give equal and opposite momenta. In any other frame the energies differ and the directions are not opposite, and the two departures are tied together by the pair’s momentum. The deviation was measured almost as soon as coincidence counting made it possible: in 1950 Stanley DeBenedetti and colleagues swung one detector about the line through a source and its partner and found the counting rate fall off over a fraction of a degree, a width that told them the positrons were annihilating with electrons of a few kilo-electronvolts’ momentum and not with electrons at rest.
A pair that is moving
Take the pair’s total energy and momentum and form the invariant that the invariant that survives a boost showed is the same in every frame — the pair’s mass. For an electron and a positron with small kinetic energies it is very nearly . In the frame where the pair is at rest the two photons leave back to back, each with energy , in a direction that the decay chooses at random. Now view that decay from the laboratory, where the pair moves with momentum . A photon that left at right angles to the motion is tilted towards it, by an angle whose tangent is , and its partner is tilted the same way. The opening angle between them falls short of 180 degrees by
where is the component of the pair’s momentum across the photons’ line.
The electron’s momentum is what electrons in molecules have: a few kilo-electronvolts per , the momentum that goes with being confined to a region an ångström across. In water the measured distribution of the deviation has a full width at half maximum of about 0.54 degrees, which corresponds to 4.8 keV/c across the line. An inner-shell electron of oxygen, held in a region eight times smaller, has a momentum of about thirty keV/c and would tilt the photons by three degrees; a positron, repelled by the nucleus, rarely gets close enough to annihilate with one.
The size of that momentum needs no detailed theory. An electron bound in a molecule is confined to a region about an ångström across, and sharpness has to be paid for: confinement to a distance costs a momentum spread of order , about two keV/c for an ångström. Equivalently, valence electrons have kinetic energies of ten electron-volts or so, and a kinetic energy of ten electron-volts for an electron is a momentum of about three keV/c. Either way the electrons the positron meets are moving at about one per cent of the speed of light, and the pair they form with a stationary positron moves at half that.
The photons’ line is therefore not the line through the annihilation point. It is a line through a point near it, tilted by a fraction of a degree, and the reconstruction has no way to know which way it was tilted.
A first-order effect at half a per cent of c
It may seem odd to call this a relativistic effect when the pair moves at half a per cent of the speed of light, a speed at which relativity’s corrections to clocks and lengths are parts in a hundred thousand. But the tilt is not one of those corrections. It is first order in the speed — the same aberration that the sky that crowds into a cone found tilting the apparent directions of stars for an observer in motion, by an angle about equal to — and each photon is tilted by roughly the pair’s , so the two together miss being opposite by twice that. Massless particles make first-order effects of speed visible because they have no rest frame of their own: any motion of their source shows up directly in their directions.
The general rule is the one the cone a decay cannot leave set out for any particle decaying in flight. In the frame where the parent is at rest the products leave back to back; in a frame where it moves fast they are thrown forward into a cone, and for a neutral pion decaying into two photons at high energy, the two photons emerge almost parallel, separated by an angle of about twice one over the pion’s Lorentz factor. An annihilating electron–positron pair is the same decay at the other extreme of speed: a Lorentz factor of 1.00001, and two photons almost exactly opposed. The PET scanner lives at the slow end of a rule whose fast end is a particle physicist’s.
A wider ring sees a wider blur
How far the tilt moves the line depends on how far the detectors are from the annihilation. For an event at the centre of a ring of diameter , a tilt of θ shifts the line at the centre by about , and with the measured spread for water that is the familiar rule of 0.0022 times the ring’s diameter.
For a small-animal scanner fifteen centimetres across the blur is a third of a millimetre. For a clinical whole-body scanner eighty centimetres across it is 1.76 millimetres. The other contributions add to it: the width of the detector crystals, which sets how finely each photon’s arrival point is known; the distance the positron travels before it annihilates, about half a millimetre for fluorine-18’s low-energy positrons and several millimetres for higher-energy emitters such as rubidium-82; and errors in deciding which crystal was hit. William Moses’s widely used estimate combines them in quadrature. With today’s four-millimetre crystals a whole-body ring resolves about 3.4 millimetres and the crystals dominate. Shrink them to a millimetre and the resolution improves only to 2.4, because the tilt is now the largest term; shrink them to nothing and the tilt and the positron range still leave 2.3 millimetres.
That is why the finest PET images come from small rings rather than better detectors. A brain scanner or a dedicated breast scanner, with a ring a third the size, cuts the tilt’s contribution by three, and a small-animal scanner almost removes it. And it is why the long, total-body scanners built in the last few years gain their improvement in sensitivity, catching more of the photon pairs, rather than in sharpness: their ring is still eighty centimetres across, and the tilt is still there.
The same motion smears the energy
The pair’s momentum has a component along the photons’ line too, and that one does not tilt the line; it shifts the photons’ energies. One photon goes slightly forward with the pair’s motion and is Doppler-shifted up, the other slightly back and is shifted down, each by half the momentum component times .
So the 511 keV line from annihilation in water is not sharp: it is 2.4 keV wide, far wider than lines from nuclear transitions of similar energy, which are measured in electron-volts. A high-purity germanium detector, whose own resolution at 511 keV is about 1.3 keV, sees the broadening clearly, and the shape of the broadened line is a measurement of the electrons’ momenta along one direction. A PET scanner’s scintillation crystals have an energy resolution of about ten per cent, fifty-one kilo-electronvolts, and see none of it; they use the energy only to reject photons that have scattered in the patient and lost energy. The scanner cares about the angle, not the energy, and the angle is the part it cannot correct.
That two independent consequences of one momentum — a tilt across the line and a shift along it — appear together is a direct statement that the pair’s momentum is a vector and the two photons share it. A photon with a momentum found Compton measuring light’s momentum by the energy a photon lost in recoil; here the photons are the ones recording the recoil, in their angle and in their energy, of a pair that is moving at about half a per cent of the speed of light.
In a metal, the tilt draws the Fermi sphere
What blurs a medical image is, in a metal, a measurement. In a metal the conduction electrons fill every momentum state up to the Fermi momentum, as no two in the same state required — a sphere in momentum space, in the simplest metals, with every state inside occupied and every state outside empty. A positron thermalised in the metal annihilates with one of these electrons at random, and the pair’s momentum is that electron’s momentum.
Count photon pairs against their deviation in one direction and the count at each angle is proportional to the number of electrons with that momentum component — the area of a slice through the Fermi sphere, which shrinks as the slice moves out from the centre. The result is an inverted parabola, ending sharply at the angle that corresponds to the Fermi momentum: 6.76 milliradians, 0.39 degrees, for aluminium. The sharp edge is the edge of the full sea that Pauli blocking leaves no electron outside. In a real metal the Fermi surface is not quite a sphere — it bulges and dimples along the crystal’s directions — and measuring the angular correlation of annihilation radiation in two dimensions, for crystals cut along different axes, maps its shape. The technique, developed from the 1950s, gave some of the first Fermi surfaces of metals and alloys, including disordered alloys where other methods fail.
The quantity that limits the sharpness of a scan of a human brain and the quantity that maps the electrons of a copper crystal are the same angle. The hospital and the solid-state laboratory sit at opposite ends of one measurement: the first wants the angle to be zero, and the second wants to know exactly what it is.
The clock says where along the line
The tilt blurs the image across the line. A different measurement places the annihilation along it: the difference in the two photons’ arrival times. If one detector is hit 200 picoseconds before the other, the annihilation was three centimetres nearer to it than to the midpoint, because light travels six centimetres in that time and the difference is shared.
Current scanners resolve arrival times to two to five hundred picoseconds, which places the annihilation along the line only to within three to seven and a half centimetres — not enough to replace the reconstruction, but enough to tell it roughly where on each line to put the event, which reduces the noise in the image substantially. A timing resolution of ten picoseconds, which several groups are pursuing, would place each annihilation along its line to 1.5 millimetres directly. Across the line, the tilt would still be there: the photons’ arrival times say nothing about their angle.
The other blur, from how far the positron goes
The tilt is one of two blurs that belong to the physics rather than to the detector. The other is the positron’s range: it is born with up to 0.63 MeV of kinetic energy from fluorine-18 and must slow to rest before it annihilates, wandering a fraction of a millimetre through tissue in a zigzag path, so the annihilation point is not the decay point. For fluorine-18 the effective blur is about half a millimetre; for isotopes with more energetic positrons — gallium-68, used for imaging certain tumours, or rubidium-82, used for imaging the heart — it is one and a half to several millimetres, and it then matters more than the tilt. The two blurs have opposite remedies. A smaller ring reduces the tilt’s blur and leaves the range unchanged; a strong magnetic field along the scanner’s axis curls the positrons’ paths and shortens their range across it, which has been demonstrated in combined PET and magnetic resonance scanners and leaves the tilt unchanged. Neither remedy can touch the other blur.
Positronium, and where the positron is
In water, about a third of positrons do not annihilate with a passing electron directly but first bind to one, forming positronium — the atom with no nucleus. Para-positronium, the singlet, annihilates into two photons within about 125 picoseconds, and its own internal motion is small, so its photons are nearly back to back. Ortho-positronium lives a thousand times longer and annihilates into three photons unless, as usually happens in tissue, its positron is picked off first by an electron of a neighbouring molecule — which returns the pick-off photons to the broad distribution of the surrounding electrons. The measured spread in water is an average over all these routes. Its width is set mainly by the momenta of the electrons the positron finally meets, and it differs slightly between tissues and with temperature, which is one reason the 0.0022 rule is a rule rather than a constant.
What the kinematics leaves out
The tilt is computed for a positron exactly at rest; in fact positrons are thermalised to energies of a few hundredths of an electron-volt, whose momentum is negligible beside the electron’s. The spread is modelled as a single Gaussian of 4.8 keV/c full width, which describes water’s measured distribution well near its centre and less well in its tails. The aluminium curve uses the free-electron sphere and an illustrative core tail; real angular correlation spectra are smeared by the instrument’s resolution and by the positron’s own distribution in the crystal. The resolution formula is an empirical combination, with a factor of 1.25 for reconstruction, and a scanner’s real resolution depends on depth of interaction in the crystals, on where in the field the event happened, and on the algorithm.
Still open: whether a scanner can see the positron’s own clock
The newest scanners time photons well enough to ask a different question of them. Ortho-positronium’s lifetime in tissue — a few nanoseconds, shortened by pick-off — depends on the size of the empty spaces between molecules, which differs between healthy and diseased tissue, and a scanner that records the time between the positron’s emission (marked by a prompt gamma ray from certain isotopes) and its annihilation could make an image of positronium lifetime. The first such images were made in the early 2020s. Whether positronium imaging carries clinically useful information beyond what the tracer’s distribution already shows, and how far it can be pushed with isotopes and detectors that exist, is an open question being tested now.
Two annihilation photons are back to back only in the frame where the pair is at rest; in tissue the electron carries a few keV/c, so the pair’s motion tilts both photons by p/mc — a 0.54° spread in water — and shifts their energies by half its component along the line, a 2.4 keV-wide 511 keV line. Across an 80 cm ring the tilt is a 1.76 mm blur that no crystal can remove, so that even with infinitely fine detectors a whole-body scanner stops near 2.3 mm; in a metal the same angle is an inverted parabola ending at the Fermi momentum, 6.76 mrad in aluminium, and the hospital’s unwanted half-degree is the solid-state physicist’s map of the Fermi surface.
Part 14 of 14
This essay is one argument about Mass-energy. The others:
The objects named here
The third axis, after the field and the reading path: the things themselves, and every essay that touches each one.
AnnihilationDoppler broadeningFermi surfaceInvariant massPositronTime of flight
- The mass a stopwatch stops being able to weigh invariant mass, time of flight