Gravitational Lensing: How Mass Bends Light into Rings
Mass bends light twice as far as Newton predicted: 1.7 arcseconds at the Sun's edge. How Einstein rings, dark matter maps and H₀ all follow.
Table of Contents
The Telescope Nobody Built
In 1979 a team led by Dennis Walsh took spectra of two faint blue objects flanking the radio source 0957+561 and got an impossible answer. Both were quasars. Both sat at redshift 1.41. Their emission lines matched line for line. Two unrelated quasars six arcseconds apart do not agree to that precision, so there was only one quasar, seen twice, its light split by a massive elliptical galaxy at redshift 0.36 lying almost exactly in the way.
That is gravitational lensing, and the physics fits in three words: mass bends light. The rings, the arcs, the dark matter maps and an independent measurement of the expansion rate all come from working out exactly how much.
Einstein and the Factor of Two
Newtonian gravity already bends light, provided you are willing to treat a light corpuscle as a falling stone. Do the calculation and a ray grazing the Sun swings by 0.87 arcseconds. Einstein did essentially that calculation in 1911, using the equivalence principle and a varying speed of light rather than corpuscles, and the paper went to press with 0.83″ printed in it — a slip, since his own formula gives 0.87″.
By 1915 he had the full field equations of general relativity, and the answer changed. Newton bends only time, loosely speaking; general relativity bends space as well, and space contributes an equal share. The deflection is
α = 4GM/(bc²)
for a ray passing mass M at closest approach b, and at the solar limb that is 1.7″ — exactly twice the Newtonian figure. That factor of two is a clean, falsifiable fork between two theories of gravity, which is why Arthur Eddington’s expedition to Príncipe for the eclipse of 29 May 1919 mattered. Light from the Hyades, grazing the Sun during totality, landed where Einstein said it would.
Where the Ring Comes From
Push the geometry one step further. If a point mass sits exactly on the line of sight to a background source, there is no reason for the light to pick one side over another, so it arrives from every direction at once and the source appears as a ring. Its angular radius is
θ_E = √(4GM/c² × D_LS/(D_L D_S))
with D_L the angular diameter distance to the lens, D_S that to the source, and D_LS that from lens to source. There is one trap here that catches people constantly: angular diameter distances do not add. In an expanding universe D_LS is not D_S − D_L. For a 10¹² M☉ galaxy at redshift 0.5 lensing a source at redshift 2, subtracting gives 468 Mpc, while the correct D_LS is 1097 Mpc — more than twice as large. Do it properly and the Einstein radius is 2.03″.
Notice where the mass sits: under a square root. Make the lens ten times heavier and the ring grows by only about three. That is why galaxy-scale arcs run to an arcsecond or so and cluster-scale arcs to tens of arcseconds, instead of spanning the wild range cosmic masses do.
Three Regimes, One Equation
Nature almost never arranges a perfect ring, and what you actually see depends on how much mass is involved and how good the alignment is. Astronomers sort the results into three regimes, but they are all the same deflection formula at different strengths.
Strong lensing is the spectacular end, and the only regime where the sky behaves anything like a real optical lens. The quasar Q2237+0305, known as the Einstein Cross, sits behind the bulge of a nearby galaxy at redshift 0.0394 and appears as four images packed into 1.6 arcseconds. The cluster Abell 1689 at redshift 0.183 does it wholesale: Hubble imaging picked out 114 separate images belonging to 34 background galaxies, smeared into arcs that curve around the cluster core.
Weak lensing is the quiet end, and far more useful statistically. Most background galaxies are not multiply imaged; they are merely stretched, typically by one or two percent, tangentially around the mass. Galaxies are lumpy and randomly oriented to begin with, so a single one tells you nothing. Average the apparent orientations of a hundred thousand of them and the shared tangential stretch emerges from the noise.
Microlensing is lensing by a single star. A solar-mass lens halfway across the galaxy gives θ_E of about a milliarcsecond, far too small to resolve, so you never see two images — only the sum of their brightnesses. The background star swells and fades over weeks as the lens drifts past, and a planet orbiting that lens adds a sharp extra blip. Unlike every other planet-hunting method, this one works on worlds that emit nothing at all.
Weighing What Refuses to Shine
Here is why lensing became indispensable rather than merely beautiful. The deflection depends on mass and nothing else — not on whether that mass radiates, absorbs or has a temperature. Every other way of weighing a cluster leans on assumptions about light.
So when the lensing mass of a cluster turns out to be five or six times the mass in its stars and hot gas combined, that is not an inference drawn from a model of how clusters glow. It is a measurement, and it is one of the firmest pillars under the case for dark matter. Shear maps go further: since the stretch of each galaxy tracks the projected mass along the sight line, a shear field inverts into a two-dimensional mass map. In colliding clusters those maps show the dark matter carrying straight through while the gas piles up and shocks — exactly what you expect of something that feels gravity and almost nothing else. The same technique, applied over billions of years of cosmic history, tracks how fast structure grew and feeds back into constraints on dark energy.
A Clock Made of Bent Light
In 1964 Sjur Refsdal published a short paper in Monthly Notices of the Royal Astronomical Society, volume 128, page 307, making a point that took thirty years to become practical. The images of a lensed source do not take the same route: one path is geometrically longer, one dips deeper into the gravitational potential and is slowed more. A flicker in the source therefore does not reach us at the same moment in every image.
The Twin Quasar flickers in one image roughly 420 days before the other. That delay is a length measured in physical units rather than rungs of a distance ladder, which makes it a direct handle on the expansion rate. The hard part is never the timing; it is modelling the lens mass well enough to convert the delay. The TDCOSMO collaboration’s 2025 analysis combined eight time-delay quasars with fifteen galaxy-galaxy lenses and stellar velocity dispersions from JWST, Keck and the VLT to get H₀ = 71.6 (+3.9/−3.3) km/s/Mpc. That sits between the cosmic microwave background value near 67 and the Cepheid value near 73, and it arrives by a route sharing no systematic errors with either, which is what the argument over how fast the universe is expanding needs.
The Magnifying Glass Nobody Paid For
A lens that spreads a source into arcs also concentrates its light, and magnification is free aperture. A cluster with μ ≈ 10 makes a telescope behave as though it had collected ten times more photons, which is why JWST keeps pointing at massive clusters: the galaxies of real interest sit behind them. The extreme case is a point in the Sunrise Arc behind the cluster WHL0137-08, nicknamed Earendel, at redshift 6.2, magnified by a factor of thousands — though recent JWST modelling suggests it may be a small cluster of stars rather than the single star first claimed. Either way, no mirror humans could build would have reached it.
What the Sky Was Hiding
The settled part of gravitational lensing is the physics; the argument over the factor of two ended in 1919. What matters now is that a consequence nobody asked for turned into instrumentation. The universe scattered lenses across the sky at random, aimed them nowhere in particular, and left them running. Learning to read them handed astronomers a mass detector blind to light, a magnifier with no aperture limit, and a clock calibrated in days that measures billions of light years. Multimessenger astronomy found the same lesson in neutron star mergers: the best instruments are often the ones already switched on.
Run it yourself: the black hole lensing simulation bends a background starfield around a compact mass so you can see the ring form and break.
Frequently Asked Questions
What is gravitational lensing?
Gravitational lensing is the bending of light by mass. In general relativity a massive body curves the spacetime around it, and light taking the straightest available path through that curved region arrives travelling in a different direction than it set out. A galaxy or galaxy cluster sitting between you and something more distant therefore works as a crude lens: it can magnify the background source, stretch it into arcs, or split it into several separate images of the same object. For a ray passing a mass M at closest approach b the deflection is α = 4GM/(bc²), exactly twice what Newtonian gravity gives. At the edge of the Sun that comes to 1.7″, the number Arthur Eddington sailed to Príncipe to measure during the eclipse of 29 May 1919. Lensing is now a routine instrument rather than a curiosity, used to weigh dark matter, to magnify galaxies too faint to see unaided, and to measure how fast the universe is expanding.
How do you calculate the Einstein radius of a lens?
The Einstein radius is the angular radius of the ring you would see if a point mass sat exactly on the sight line to a background source. It is θ_E = √(4GM/c² × D_LS/(D_L D_S)), where D_L is the angular diameter distance to the lens, D_S the distance to the source and D_LS the distance from lens to source. The trap is that angular diameter distances are not additive: D_LS is not D_S − D_L. For a 10¹² M☉ lens at redshift 0.5 with a source at redshift 2, the correct D_LS is 1097 Mpc rather than the 468 Mpc you get by subtracting, and the Einstein radius comes out at 2.03″. Because the mass sits under a square root, a ten-times heavier lens produces a ring only about three times wider, which is why cluster-scale arcs are tens of arcseconds across and not thousands.
What is the difference between strong, weak and microlensing?
They are one piece of physics at three strengths. Strong lensing occurs when the mass is large enough and the alignment good enough to make multiple images of a single source, giant arcs, or a complete Einstein ring; the Twin Quasar and the Einstein Cross are the textbook cases. Weak lensing is the regime where no extra image appears and the only effect is a slight stretching of each background galaxy, usually by a percent or two. Galaxies have untidy shapes to begin with, so weak lensing is measured statistically, by averaging the apparent orientations of thousands of them, and that is how most dark matter maps are built. Microlensing is lensing by a single star or compact object, where the images sit far too close together to separate, so the only observable is a smooth temporary brightening of the background star as the lens drifts across it. That brightening is how hundreds of exoplanets and free-floating objects have been found.
Can gravitational lensing measure the Hubble constant?
Yes, and it does so without a distance ladder. Sjur Refsdal pointed out in 1964, in Monthly Notices of the Royal Astronomical Society volume 128, page 307, that the multiple images of a variable source reach us along paths of different length and through different depths of gravitational potential, so a flicker in the source shows up in one image before the other. Measure that delay, model the mass of the lens, and a lag in days converts into an absolute distance, which fixes the Hubble constant. The difficulty is the mass model, not the timing. The current best result from this route is the TDCOSMO collaboration analysis of 2025, which combined eight time-delay quasars with fifteen galaxy-galaxy lenses and stellar velocity dispersions from JWST, Keck and the VLT to give H₀ = 71.6 km/s/Mpc, with an uncertainty of +3.9 and −3.3.