Entanglement Isn't Spooky. It's Shared.

Here's a trick you can do with a pair of gloves.

Take one left glove and one right glove. Without looking, seal one in each of two boxes. Mail one box to a friend in another city and keep the other.

Now open yours. It's a left glove. And at that instant you know your friend's box holds a right one.

You didn't phone anybody. Nothing travelled between the boxes. The answer was packed in advance, and opening your box only told you which half of the pair you'd been sent.

Nobody calls that spooky. It's gloves.

Here's the part that does get called spooky. Physicists have a test where the best a pair of gloves can do is a score of 2. Pairs of entangled particles beat it. Test after test since the early 1970s has found that they do. Quantum physics puts their ceiling at 2.83. (The limit on gloves is called a Bell inequality. In 2022 the Nobel Prize in Physics went to Alain Aspect, John Clauser and Anton Zeilinger for experiments with entangled photons that established its violation.¹)

Einstein never made peace with it. In 1947 he wrote to his friend Max Born that he couldn't seriously believe in quantum physics, because it couldn't be squared with a physics "free from spooky actions at a distance."² The phrase stuck. It's still how most people first hear about entanglement.

So what are entangled particles doing that gloves aren't? A wire? A signal? Something faster than light?

None of those. Here's the answer.

The short answer

Entangled particles share one record. It's the permanent record reality keeps of every event (more on it in a minute).

It isn't two separate things joined by a hidden string, and nothing flies between them. There's one record, and there are two places to act on it. A measurement on either side doesn't just read the record. It moves it. And the move shows up only when the two sides' answers are put next to each other.

Here's the whole answer in five steps. The rest of this post takes them one at a time and shows how we know each.

1. A plan has a ceiling. Two far-apart friends play a game for many rounds. Whatever plan they make beforehand, the best they can do is win three rounds in four. Physicists score that as 2.
2. A record you only read is still a plan. If the friends share one record and each only looks at it, nothing changes: three rounds in four at most, a score of 2.
3. A record you move beats every plan. If each friend's question moves the record, the way the record's own basic step does, they win about 85 rounds in 100. The score is exactly 2√2, about 2.83, the most quantum physics allows.
4. No message gets through. Each friend's answers, taken alone, look like fair coin flips, whatever the other friend does. The pattern shows up only when the two lists are put side by side.
5. The pair is sharp. Each half is a blur. The pair as a whole is in one exact state. Each particle on its own isn't.

Why this matters: the usual telling of entanglement makes it sound like magic, a particle here instantly changing a particle there. Neither half of that is true. Nothing is sent, and the other side's answers, taken on their own, never change. What's true is simpler and stranger: two particles can be two ends of one thing. That's how a pair can beat every plan and still never carry a word.

What you need from Part 1

This post stands on one idea from Part 1 of this series, so here it is in a minute.³

Reality keeps a permanent record of every event, and nothing in it is ever erased. We call it the ledger, like a bookkeeper's: entries go in one after another, and they stay. (Another post follows that rule into your trash can: The Universe Has No Delete Key.)

The ledger writes in rows of marks. Each mark is full or empty, like the tokens in Part 1. I'll write full as 1 and empty as 0. Flipping a mark, 0 to 1 or 1 to 0, is the ledger's own basic move. (It's the same flip as in Part 1.)

If the ledger had a personality, it would be unflappable, thorough, and armed with one answer to everything that has ever happened: Noted. When it has a face, I'll call it the Ledger, with a capital L. The face is a drawing. What it stands for isn't: the actual, permanent record of everything that happens.

Meet Yeah-But, too: the voice in the back of your head that has never once said wow. It has questions.

Here's today's route. The Ledger will keep the map, and I'll stop after each step to zoom out.

Step 1: A plan has a ceiling

Two far-apart friends who follow a plan made beforehand win at most three rounds in four. Here's the game, and how we know.

Alice and Bob sit in two soundproof booths. Each round, each of them is dealt a card, 1 or 2, and answers yes or no. Once the cards are out, they can't talk. They win the round if their answers match. There's one exception: if both of them were dealt card 2, they win by answering differently.

The booths are a drawing. In the real test, Alice and Bob are two measuring devices, each holding one particle of a pair. The card says which measurement to make, and the answer is what it shows. Distance and timing do the soundproofing: the two sides are far enough apart, and the dealing and answering are quick enough, that each friend has answered before light from the other friend's card can arrive. Physicists call the game the CHSH test, after the four who set it up in 1969: Clauser, Horne, Shimony and Holt. It builds on an argument John Bell published in 1964.⁴

Before the game starts, Alice and Bob can agree on a plan. Say they agree: "Always say yes." Check all four pairs of cards:

Three pairs out of four. The cards come out at random, so each pair turns up a quarter of the time. That's 75 rounds in 100.

Can a cleverer plan win all four? No, and a loop shows why. A plan is a fixed answer for each card: Alice's answer to card 1 and to card 2, and Bob's answer to card 1 and to card 2. To win every pair of cards, those four answers have to satisfy the four links in this loop.

Walk around the loop. Three links say "same" and one says "different." Start with Alice's card 1 answer and follow two "same" links: Alice's card 2 answer comes out equal to it. The "different" link then says Bob's card 2 answer is the opposite of that. But the last link says Bob's card 2 answer matches Alice's card 1 answer, the very answer we started with. You come back flipped. Every plan breaks at least one link, and the best plans break exactly one.

What if they flip a coin to choose their plan? A random choice among plans is an average of plans, and an average can't beat the best plan in it. So three rounds in four is the ceiling for every plan, however it's chosen.

The score. Physicists don't count rounds. They score how strongly the friends agree on each pair of cards: +1 means they always agree, −1 means they always disagree, and 0 means no link at all, like two unrelated coin flips. Add the scores for the first three pairs of cards and subtract the score for the last pair (cards 2 and 2, which they win by differing). Call the total S.

"Always say yes" scores 1 + 1 + 1 − 1 = 2. No plan scores more. That limit of 2 is Bell's inequality in the form Clauser, Horne, Shimony and Holt wrote down. Each point of S is worth 12.5 rounds in 100, starting from 50 for pure guessing, so the win chance is 50 + 12.5 × S. A score of 2 means 75 rounds in 100.

That's what a pair of gloves can do. Gloves are a plan.

Zoom out. Step 1 is done. A plan, however it's chosen, wins at most three rounds in four: a score of 2. Next, and why: the friends can do more than carry a plan. They can share one record. Does that beat a plan?

Step 2: A record you only read is still a plan

A shared record doesn't beat a plan if the friends only read it. Here's the record, and why.

Take four marks and write them as one row, with a bar between the left pair and the right pair: 00|00. The record isn't one such row. It's four rows at once, each with the same weight, and in every row the two halves match:

00|00, 01|01, 10|10 and 11|11.

This isn't two copies of anything. It's one record, with two places to act on it. Alice acts at the left end and Bob acts at the right end. In the real test, the two ends are the two particles. (Alice and Bob are the physicists' two favorite volunteers. They've sat in a great many soundproof booths and have never once complained.)

The simplest thing a friend can do with the record is look at it. A read reports something about the record, such as "Is mark 1 of my half full?", and leaves the record exactly as it was. Any yes-or-no question that works this way counts.

Reading never changes the record. So every answer either friend can be asked for was already sitting in the record before anyone looked. Take the row 00|00. It already holds Alice's answer to every card and Bob's answer to every card: that row is a plan. With reads only, the record behaves like one of its rows picked at random, with that row's weight, and the row is a plan. A random pick of plans can't beat three rounds in four.

That holds for every shared record and every way of reading it, not only this one. A record that's only read is a plan, so the Bell limit from Step 1 applies: it never scores above 2, and 2 is reached.

Yeah-But has been waiting. "So if my friend and I share a record and we both just look at it, that's gloves?"

That's gloves. A pair of gloves is the simplest shared record there is.

Zoom out. Step 2 is done. A shared record that's only read is a plan written down: at most 2, and 2 is reached. Next, and why: the friends' questions can do something a read never does. They can move the record.

Step 3: A record you move beats every plan

When each friend's question moves the shared record, the way the ledger's own step does, the score reaches exactly 2√2, about 2.83. That's about 85 rounds in 100, and no plan comes close. Here's how it works, with the numbers.

The ledger's move. The ledger's own move is a flip: pick a mark and turn it over. In the ledger a flip carries a sign, +1 or −1. It's +1 if an even number of marks before it (in its own half) are full, and −1 if an odd number are. That's the ledger's own parity rule.

Here's what the sign does when two moves meet. Start from the half 0 0 and make both flips, mark 1 and mark 2, in either order:

The same row comes out. The sign doesn't. The order of two moves matters. (The Ledger notes both orders. It notes everything.)

A read never does this: reading mark 1 and then mark 2 gives the same as reading 2 and then 1. Because the order of moves matters, no row of the record can hold the answers to both of Alice's questions in advance. If a row held both answers, asking them in either order would give the same two answers. A plan is a list of answers that are all there from the start. A move breaks that. Read, the record behaves like one row picked at random. Moved, it behaves like all four rows at once.

Both kinds of question, reads and moves, are yes-or-no questions that undo themselves: do a move twice and you're back where you started. So the yes-or-no form isn't what buys the extra. The move is.

The friends' moves. Alice's card picks her question. Card 1 asks the flip-mark-1 question on her half. Card 2 asks the flip-mark-2 question. Bob's card picks his, and his questions are even blends: equal parts of his flip 1 and flip 2. Card 1 asks the blend with the two added. Card 2 asks the blend with the second subtracted. A blend of two moves is itself a proper move that undoes itself, and that works only because the order of the two moves matters. Without the sign, the blend wouldn't undo itself.

How strongly do their answers agree? For each pair of cards, it's the overlap between the record after Alice's move and the record after Bob's move. The overlap is 1 if they're the same record, 0 if they have no row in common, and −1 if they're the same record with its sign turned over.

Look at the rows. Alice flips her mark 1, and the record becomes four new rows. Bob flips his mark 1, and the record becomes the same four rows. A move at either end lands on the same record. That's what "one record, two places to act" means. Overlap 1. (Each row also carries a sign, and the signs match on both sides, so I've left them out of the drawing.)

Bob flips his mark 2 instead, and the record becomes four different rows, with none in common with Alice's. Overlap 0.

Now the trick. Bob's questions aren't plain flips, so his records aren't the plain flip records. They're even blends of them.

I'll draw each record as an arrow of length 1 on a flat sheet of paper. The arrows are a drawing of the real thing. The overlaps they carry are exact. The overlap of two arrows is the length of the shadow one casts on the other: 1 if they point the same way, 0 at right angles, 0.707 at 45 degrees (the cosine of 45 degrees, if you know your trig). Records with no row in common point at right angles. So Alice's two records, from flip 1 and flip 2, sit at right angles. Bob's two records are even blends of hers, so they sit 45 degrees off.

Read the four overlaps off the arrows. Alice 1 with Bob 1: 45 degrees, 0.707. Alice 1 with Bob 2: 45 degrees, 0.707. Alice 2 with Bob 1: 45 degrees, 0.707. Alice 2 with Bob 2: 135 degrees, −0.707 (the shadow points backwards, so it counts as negative). Now score it the way the game does, adding the first three and subtracting the last:

0.707 + 0.707 + 0.707 − (−0.707) = 4 × 0.707 = 2.83.

That's exactly 2√2 (the square root of 2 is 1.414, and twice that is 2.83). The win chance is 50 + 12.5 × 2.83, or 85.4 rounds in 100.

Remember the loop? A plan has to break one link completely. Here, the loss is shared out: the friends win every pair of cards about 85 times in 100.

This is one working example, built on the ledger's own rows. It needs at least two marks in each half, and it works for any larger number. It isn't a claim that every lab's equipment is built this way. It shows that a shared record moved by the ledger's own flips beats every plan, and by exactly the quantum amount.

It stops there. However you turn arrows of length 1, you can't score more than 2.83. Boris Tsirelson proved that ceiling for quantum physics in 1980.⁵ The ledger's count lands on the same number, so nothing here disagrees with quantum physics. Quantum physics writes the pair as one joint state. The ledger says what that state is: one record, with two places to act on it.

Does anything allow a score above 2.83? If the only rule is "no messages," yes. In 1994 Sandu Popescu and Daniel Rohrlich described a box that obeys that rule and scores a perfect 4.⁶ It wins every round and sends nothing. Mix it with an ordinary entangled pair and you reach every score between 2.83 and 4, all without a message. Quantum physics rules the box itself out. So do the arrows: to score 4, the loop would need arrows that point the same way three times and the opposite way once, and arrows of length 1 can't do that. So "no messages" isn't what stops the score at 2.83. For friends acting on one record, the shape of what's being scored does: overlaps of records, which are arrows of length 1.

What real tests find. Real equipment is noisy, so measured scores come in under the ceiling. In 2015 a team at Delft University of Technology entangled two electron spins in diamonds in two labs 1.3 kilometres apart on opposite sides of the campus. Distance and timing played the part of the soundproof booths: each side picked its card and recorded its answer before light from the other side's card had time to arrive. Over 245 rounds they scored 2.42, give or take 0.20.⁷ That's above the best a plan can do by about two margins of error. A plan would produce a score that high by luck about 4 times in 100.

Zoom out. Step 3 is done. A shared record moved by the ledger's own flips scores exactly 2√2, about 2.83: 85 rounds in 100, with the loss shared evenly across the four pairs of cards. No arrangement of arrows scores more, and a plan tops out at 75. Next, and why: a score that high sounds like Alice's answer has to reach Bob somehow. Does anything cross the gap?

Step 4: No message gets through

Whatever Bob does, Alice's answers, taken on their own, look like fair coin flips. The match between the two lists shows up only when they're put side by side. So Bob can't use his cards to send Alice a word.

Here are two lists from eight rounds in which both friends were dealt card 1.

Each list on its own is a perfectly fair coin: you can't tell it from flipping one. Side by side, 7 of 8 match. (With a score of 2.83 the friends match about 85 times in 100 on this pair of cards, so 7 of 8 is a typical run.)

Why Alice's list is a coin. Her move turns the record into a different record, with none of the original rows in it. So by itself it reads nothing about where the record started, and yes and no come up equally often. The same holds for Bob's list, whichever card he's dealt.

Why Bob can't steer it. The two friends' moves never get in each other's way. Alice's moves touch only her half, and count only the marks in her half. Bob's do the same with his. Do Alice's move first and Bob's second, or Bob's first and Alice's second, and the record comes out the same. If Bob's choice of card changed Alice's odds, the order would matter. It doesn't. This is the same reason quantum physics gives for why entanglement never carries a message: acts on two separate particles always come out the same in either order.

Yeah-But again. "But Alice's move changes the record, and the record is shared. Doesn't Bob's end feel it?"

The record changes. But nothing Bob can ask, taken on its own, has different odds afterward. The change shows up only in the pairs: how Alice's answers line up with Bob's. That's why it can win a game and can't send a word.

What would break it. No message isn't automatic. It depends on how odds are counted. Each branch of a record has a strength, and its chance is its strength squared. Here's one counting rule that fails.

Take a record with two equal branches, "yes" and "no," each with strength 0.707. Each has chance 0.707² = 0.5. Now Bob cuts the "no" branch into two equal pieces. (He can, because each of Alice's branches is tied to a different state of his side, and his cut lands on only one of them.) Each piece has strength 0.5, and 0.5² + 0.5² = 0.5. The "no" branch still has chance 0.5. Cutting changed nothing.

Count with fourth powers instead. Before the cut, "yes" counts 0.707⁴ = 0.25 and "no" counts 0.25, so each has chance 1/2. After the cut, "yes" still counts 0.25, but the two pieces count 0.5⁴ = 0.0625 each, 0.125 together. Now "yes" has chance 0.25 ÷ (0.25 + 0.125) = 2/3. Bob's cut moved Alice's odds from 1/2 to 2/3, and he has a phone line. Squares are the counting rule where cutting a branch never changes its total. That's what keeps the line dead.

Zoom out. Step 4 is done. Nothing crosses the gap. Each side's answers are fair coin flips on their own, whatever the other side does, because the two friends' moves never get in each other's way and odds are counted by squares. The score of 2.83 lives entirely in the pairs. Next, and why: if each half on its own is a coin flip, what is the pair?

Step 5: The pair is sharp. Each half is a blur.

When two particles are entangled, the pair is in one exact state, and each particle on its own isn't.

Look at the rows again. The pair's record is exact: four rows, equal weight, the halves matching in every one. Now ask about Alice's half alone. Is mark 1 full? In two rows of the four it is, and in two it isn't. A coin flip. Her half, taken alone, has no preferred row at all: all four have the same chance. That's a perfect blur. So is Bob's half. And yet the pair is exact. Put two blurs together the right way and you get one sharp record.

That holds well beyond this example. For a pair in one exact state, any amount of entanglement at all makes each half uncertain on its own: each half carries positive entropy, which is the physicists' measure of uncertainty. If the pair is two separate things, each half is exact on its own. Schrödinger described this strangeness in 1935: you can know everything there is to know about two systems together, and have lost all knowledge of either one taken separately.⁸

Compare the gloves. With gloves, the parts are exact and the whole is uncertain: each glove is definitely a left or a right, and what's uncertain is which box got which. With entanglement it's reversed. The whole is exact, and the parts are uncertain. That reversal is what "one record" means. The record is exact. It belongs to the pair. And each half, alone, has no exact record of its own.

Yeah-But has one last question. "Fine. But how do two particles end up sharing a record in the first place?"

That's the one step still being worked out: the exact rule by which two separately made particles come to share one record.

The whole trip, in one breath

Two friends in two booths play a game. A plan, like a pair of gloves, wins three rounds in four, a score of 2, and no plan does better, however it's chosen. A shared record that's only read is a plan written down, so it stops there too. A shared record that each friend's question moves, with the ledger's own flips and its sign, wins about 85 rounds in 100, ten more in every hundred than the best plan: a score of exactly 2√2, about 2.83, the ceiling quantum physics has always had. Nothing crosses the gap. Each friend's list, on its own, is a fair coin flip, because the two friends' moves never get in each other's way, and the pattern lives only in the pairs. And the pair is exact while each half is a blur.

Entanglement isn't two things with a secret. It's one thing with two ends. It isn't spooky. It's shared.

(The Ledger would like it noted that it has only ever been one book. It has never once had to call itself to check.)