Could You Send a Message With Entanglement?
The Link Is Real. Nothing Rides On It.
Make me a pair of entangled particles and I will carry one of them to Mars. You keep the other. Measure yours and mine answers at once, with no delay at any distance: you have said so yourself, and the experiments say so. That is a channel. I am not greedy about it. One bit, faster than light, once. Give me the protocol.
The answer is no, and what makes it worth reading is how hard the no is to reach. The first two walls agree with the asker completely. The correlation is real, no experiment has ever found a delay in it, and it cannot be explained by anything the particles carried away from the source. Every belief the request holds about the physics is correct. What fails is a single step the request never states out loud, and it fails to a theorem rather than to any difficulty of engineering. Then, at the end, something stranger: the theory could have been more correlated than it is and still refused you the bit, and why it stops where it does is not settled: several principles have been proposed, and none is agreed.
What Physics Says, Wall by Wall
The correlation is real, and nobody has found a delay in it
Make two particles together in the right way and their measurements are locked to each other. Ask both the same question and you always get opposite answers, however far apart the two of you are and whatever the delay between the questions. This is not a claim about what some equation says. It is one of the most carefully checked facts in physics.
Three groups closed the last objections at once in 2015. One at Delft put the two detectors 1.3 kilometers apart. One in Vienna and one in Boulder used detectors efficient enough that no sample of the pairs could be quietly thrown away. And all three chose their measurement directions too late for any influence traveling at light speed to have arranged the outcome. Nothing was left to argue about.
The word instant deserves care, and the honest form is this: no delay has ever been detected, and the floor on how fast any hypothetical influence would have to travel is enormous. An experiment in Geneva in 2008 timed the two ends against each other for a full day and came out with a lower bound of ten thousand times the speed of light. So the request is not being told the effect is slow. It is not slow.
And it is not two gloves in two boxes
The obvious way to deflate all of this is to say the answers were written on the particles before they left. Two gloves are packed into two boxes and shipped to two cities. Open one, find a left glove, and you know the other box instantly, at any distance, with no influence passing between them. Nothing mysterious, and certainly no channel.
John Bell showed in 1964 that this deflation is testable, which is the single cleverest thing anyone has done in this subject. Suppose each particle really does carry its answers to every question you might ask. Then, across three possible questions, the fractions of times the two sides agree have to obey an inequality. The inequality is arithmetic about groups of objects. It assumes nothing about physics, and it holds for gloves, for coins, for anything at all that arrives with its answers already fixed.
Quantum mechanics predicts a violation of it, and the violation is measured. The counting bound is 2, quantum mechanics allows up to 2.83, and the loophole-free experiments came out near 2.4. So the gloves picture is not merely unfashionable. It is ruled out by experiment. Whatever is happening, the answers were not sitting on the particles waiting to be read.
Everything above is granted, and here the request dies
Now actually try to send the bit. You are on Mars with your half of the pair. The only thing you control is your own measurement: which direction you ask about, or whether you ask at all. So agree in advance that one direction means one and another means zero, and start sending.
Look at what arrives here. My particle gives me up or down, and it gives me each of them half the time. Turn your dial to the other direction and it is still half and half. Turn it anywhere in between and it is still half and half. Put your particle in a drawer and never look at it, and my results are still half and half, with the same randomness and the same statistics as in every other case.
There is no measurement on my side that tells the difference. Not a cleverer detector, not a longer run, not any amount of processing. Collect a billion results and their distribution is the same one. This is the no-communication theorem, and it is not a remark about how hard the experiment would be. It falls out of the structure of the theory in a few lines: what I can observe depends only on my half of the state, and your choice of direction does not touch my half.
The correlation is still there. It is simply not anywhere. It exists in the comparison of two lists, and a comparison requires both lists in one place, and getting them into one place means sending an ordinary message at the speed of light or slower. The link cannot carry the bit because the link is not a thing that carries. It is an agreement between two records that only becomes visible once somebody has carried the records.
The picture below is that sentence as a shape. The dial sets the angle between your measuring direction and mine. The left bar is how often you get up, the middle bar is how often I get up, and the right bar is how often our two lists agree once one of them has been carried to the other. Turn the dial through its whole range. The right bar sweeps from never to always. The other two do not move at all, and there is no setting, including refusing to measure, that makes them move.
You cannot copy your way around it
The standard escape is to say the problem is sample size. Take my particle, make a million copies of it, measure them all, and read off the statistics before you have touched yours. Then have you touch yours, copy mine again, and compare. Two different statistics, one bit, done.
Copying is the step that is not available. An unknown quantum state cannot be copied, and the proof is three lines long. Any copying operation would have to be linear, because everything in the theory is, and a linear operation applied to a superposition of two states gives you something that is not two copies of that superposition. There is no device to be invented here. The operation is not the sort of thing the mathematics contains.
That wall is load-bearing in a commercial sense as well as a theoretical one. Quantum key distribution is secure precisely because an eavesdropper cannot copy what passes: any attempt to look shows up as errors in the key. Banks and governments are already buying equipment that rests on it. It is not an unexamined corner.
Nor can you use the collapse itself as a bell
The next escape does not encode the bit in which direction you measure, but in whether you measure. Measure, and my half is supposed to collapse to a definite value. Leave it alone, and my half stays entangled with yours. Those sound like two different objects. Surely they behave differently under test?
They do not, from where I stand. The thing I can experiment on is my particle considered by itself, with yours ignored, because yours is on Mars and I have no access to it. That object is the same in both cases: a completely random mixture, half up and half down. Whether the randomness arrived because you measured or because we are still entangled is not a property my apparatus can reach, since both give the same description of my particle alone.
This is the sharpest form of the refusal, because it needs no experiment to check. The two situations are described by the same object. Things that are equal cannot be told apart by any test, now or later, by anyone.
The theory could have been more nonlocal, and it is not
Here is the part that should be strange even to a reader who has accepted everything above. Turn the question around. How correlated could any theory be, in principle, while still refusing to let you signal?
Sandu Popescu and Daniel Rohrlich answered that in 1994 by inventing a device that does not exist. It is a box in two halves, one for each of you. Each half takes an input and returns a random output, and the outputs are correlated so strongly that they push the counting bound from wall two all the way to 4, the highest any correlation of that kind can go. And it still does not let you signal. Each half is random no matter what the other half is asked.
So there was room above quantum mechanics. Nature could have been more tightly correlated than it is, all the way to the ceiling, and the ban on messages would have survived untouched. Instead it stops at 2.83, comfortably short. Why it stops there is not settled: several principles have been proposed to derive the number, and none is agreed to be the reason. The candidates are respectable and none of them has won: that anything stronger would make certain communication problems suspiciously easy, or would break a principle about how much information a measurement can reveal.
The picture worth taking away is not that quantum mechanics scrapes past the speed limit. It is that quantum mechanics stops well short of it, for a reason that is still an open question.
Two numbers close together on a page are not close together on a ruler, so here is the ruler. The shaded stretch is the room nature did not take, and it is the larger of the two: from 2 up to where quantum mechanics stops is 0.83, and from there to the ceiling is 1.17. Put the marker anywhere in it and read the second line of the readout, which does not change.
The Pushback
Neither of these changes the answer. Both change what the page is entitled to claim while giving it.
Because of a detail in the middle of it that the name hides. Teleportation is real: a quantum state is reconstructed somewhere else without the object crossing the gap, and it has been done from the ground up to a satellite at ranges up to 1,400 kilometers. Look at the steps. The sender measures their particle jointly with the one being sent, and gets one of four possible results, at random. The receiver's particle is now in one of four states, and which one depends on that random result.
So the sender has to telephone the receiver and say which of the four came up. Two ordinary bits, traveling at the speed of light or slower. Until that call arrives the receiver holds something completely random and nothing has been moved. Teleportation is a way of spending entanglement to make a classical channel carry a quantum state. It has a phone call bolted through the middle of it, and it does nothing at all without one.
This is the right place to be uncomfortable, and there are exactly two honest things to say. The first is that no paradox needs resolving. A correlation is not a signal, so nothing here can be used to reach into anybody's past, and the awkward question of which measurement happened first never has to be answered, because neither side's results depend on the other's choice.
The second is that what actually happens is a question the interpretations answer differently, and they cannot all be right. On one account there is a collapse and it really is instantaneous. On another nothing collapses at all and both outcomes occur. On a third the particles had definite values the whole time and there is a genuine faster-than-light influence, permanently hidden from use by exactly the wall in the third rung above. Every one of these makes identical predictions for every experiment ever performed. Physics is telling you what you will see. It is not, here, telling you what happened.
What Would Change the Answer
One thing would break this page, and it is worth naming because it is checkable. The no-communication theorem rests on quantum mechanics being exactly linear, in the sense that the evolution of a sum of two states is the sum of their separate evolutions. Steven Weinberg looked in 1989 at what happens if you add even a small nonlinear term to the equation, as a way of testing whether the theory could be slightly wrong. Almost every version of that addition lets you signal faster than light.
So the ban and the linearity are the same fact, and linearity has been measured. Experiments on trapped beryllium ions the same year bounded any nonlinear term at about one part in 10²¹ of the energies involved. That is one of the tightest limits on anything in physics. If it ever cracks, this page is wrong and so is a great deal else.
It is worth seeing how much is stacked on this. A faster-than-light channel, combined with relativity, is not merely a fast channel. Two observers moving relative to each other disagree about which of two distant events came first, so a message that outruns light for one of them arrives before it was sent for another. Send it back and you have written to your own past. The wall in the third rung is not a rule about particles. It is the thing standing between quantum mechanics and a contradiction.
An Opinion, Dated
The walls above are physics. This box is the drafting model’s own bets, and they are scoreable.
Ninety-nine in a hundred that no experiment ever shows a message carried by entanglement alone. The remaining one is not for a clever protocol. It is for the possibility that the theory is not exactly linear, which is the only door, and it is bolted at one part in 10²¹.
One in four that by 2040 the limit of 2.83 is derived from a principle the field accepts as the reason it sits there, rather than as one more statement equivalent to it. Thirty years of good people have not managed it yet.
Three in five that entanglement is distributed directly between two satellites, with no ground station in the path, by 2035. The pieces exist and the engineering is the hard part, which is the kind of problem that usually yields.
These intuitions were distilled from human physics writing and carry its fashions. Weigh them accordingly.
Where the Request Was Right
Every factual claim in the request is correct. The correlation is real, it has no detectable delay, and it cannot be explained by anything the particles were carrying. Most confident statements about entanglement that you hear are worse than this request, not better: the common deflation, that the answers were decided at the source, is the one thing that has actually been ruled out. The asker was not being credulous. They were being more accurate than the person who tells them it is nothing special.
The single unexamined step is between a correlation you can see and a channel you can use. It looks like no step at all, which is why the request is made so often and by such good people. The distinction it hides is the difference between two lists matching and one list having been made by the other.
And entanglement does buy things, just not this. Quantum key distribution turns the no-cloning wall into security and has been run between ground stations 1,120 kilometers apart through a satellite. Superdense coding sends two bits down a channel that could otherwise carry one, provided a pair was shared in advance. Neither beats light. Both beat what the same equipment could do without entanglement. The request was reaching for the one prize that is not on the table, at a table with several others on it.
The Updated Map
The picture to give up is entanglement as a wire: two ends, and something passing down it. There is no wire, and nothing passes. What there is, is a pair of records made to match before either of them existed, and no operation on one of them that alters the other. The matching is stronger than any account in which the answers were written in advance, and weaker than the strongest matching that would still forbid a message. Both of those are measured numbers. Neither of them hands you the bit. The better image is not two ends of a cable but one object that happens to be in two places, whose halves say nothing to each other because there is nothing yet to say: the answer was never sitting on either side to be sent.
