The Parts Do Not Determine the Whole

Contents

1. The multiplication nobody mentions
2. What Schrödinger actually wrote
3. The exact test
4. What a local instrument reads
5. Nothing happens at Bob’s end
6. Bell, stated exactly
7. Entangled is not the same as non-local
8. What actually fails
9. What it will not do
10. The ledger
11. Exercises

01 The multiplication nobody mentions

Put two coins on the table. To say what the pair is doing you say what each coin is doing — heads or tails, twice — and you are finished. That is what it means for a world to be made of parts, and it is so obvious that nobody writes it down as an assumption.

Write it down as an assumption. Classically, joining two systems means taking the Cartesian product of their state spaces: list the state of the first, then the state of the second, and the dimensions add. One particle needs three numbers, two particles need six. A configuration of the pair just is a pair of configurations.

Quantum mechanically, joining two systems means the tensor product, and the dimensions multiply.

Ten qubits live in a space of 1,024 dimensions; N qubits in one of 2^N. That growth is why quantum computers are interesting. It is also why the parts go missing.

A vector in ℋ_A ⊗ ℋ_B can be a product — |ψ⟩ ⊗ |φ⟩, this state for A and that state for B — but almost none of them are. Count the room. A pure state of N qubits has 2·2^N − 2 real parameters once normalisation and overall phase are quotiented away. The ones that factor into a state per qubit have 2N. Everything else — which is to say essentially everything — is entangled.

Figure 01

How much of the space describes things that have parts

Qubits	All pure states (real dimensions)	States that factor (real dimensions)	Fraction of the directions available
N = 1	2	2	100%
N = 2	6	4	67%
N = 3	14	6	43%
N = 5	62	10	16%
N = 10	2,046	20	0.98%
N = 20	2,097,150	40	0.0019%

• the space of states the joint system can be in
• the sliver of it that has a state for each part

Here is the inversion this series keeps making, and it has rarely been cleaner. Entanglement is not a phenomenon bolted onto quantum mechanics that needs a mechanism. It is what the tensor product is. The thing needing explanation is the opposite one: why anything is ever separable — why the world presents itself as a collection of objects each carrying a state of its own.

There is an answer, and it reverses the usual picture. Systems look separable when they have leaked their entanglement into an environment nobody is tracking. Decoherence does not remove entanglement; it spreads it, into a bath of air molecules and photons and lattice vibrations, and the piece you are still looking at then behaves like an object with a state. Classicality is not the absence of entanglement. It is entanglement filed somewhere you are not looking.

Do not ask why particles become entangled. Ask why anything is ever separable. The mathematics says that is the rare case, and the rare case is the one that needs a mechanism.

02 What Schrödinger actually wrote

Three of the most-quoted lines in this subject are wrong in the form the popular record carries them. This audience will trust me because I am careful with texts, so let me be careful with these.

Schrödinger, 1935. The famous sentence is usually printed as a flat declarative. What he wrote is first person, hedged, and refers to a situation he has just described — two systems interact, then separate, and afterwards “they can no longer be described in the same way as before, viz. by endowing each of them with a representative of its own.”

I would not call that one but rather the characteristic trait of quantum mechanics, the one that enforces its entire departure from classical lines of thought.

Schrödinger · Proc. Camb. Phil. Soc. 31, 555 · 1935

Notice the antecedent of “that.” It is not a spooky link between two systems. It is the failure of the description to come apart — the impossibility of endowing each with a representative of its own. Schrödinger’s claim, in 1935, is the claim in the title of this lesson. (The English word appears in that paper; the German Verschränkung follows weeks later, in the three-part essay with the cat in it.)

EPR, 1935. The Einstein–Podolsky–Rosen paper uses position and momentum. There is no spin in it anywhere. The two-particle spin-½ setup that everyone teaches, and that half the literature calls “the EPR experiment,” is David Bohm’s reformulation from his 1951 textbook. It is a better argument, and it is not the one Einstein made.

“Spooky action at a distance.” Not in EPR, and not in any published Einstein paper. It comes from a private letter to Max Born of 3 March 1947, it is plural in the German — spukhafte Fernwirkungen — and it is a negative construction. Einstein is naming what he wants physics to be free from, not a thing he thinks is there. Every popular treatment inverts that.

03 The exact test

“Cannot be written as a product” sounds hard to check — you would have to try all the products and fail. It is not hard, because of a theorem that is the singular value decomposition wearing a physicist’s hat.

Every pure state of a bipartite system has this form, and the test is a single integer. The number of non-zero λ_i is the Schmidt rank: rank 1 means product, rank above 1 means entangled. Nothing to search.

Two pieces of precision that get skipped constantly. The bases are chosen for that state — you find them by diagonalising, and a different state gets different ones. And the theorem is bipartite only. There is no Schmidt decomposition for three parts, which is not a technical gap but a fact about the world: three qubits admit genuinely different kinds of entanglement, and no single number classifies them.

One definition to hold carefully. For pure states, “separable” and “product” mean the same thing. For mixed states they come apart: a mixed state is separable if it is some probability mixture of products, ∑_i p_i ρ_i^A ⊗ ρ_i^B, which is strictly broader than being one. There is no mixed-state Schmidt theorem, and deciding separability in general is NP-hard. Section 07 lives entirely in that gap.

04 What a local instrument reads

Part three turned on one question: what does the instrument actually couple to? Ask it again here. Alice holds one particle of an entangled pair. What can her apparatus read? Exactly one object — the reduced density matrix, got by tracing out the half she does not have. Every probability of every outcome available to her is a function of that operator and nothing else.

Sit with that. I/2 is the unique maximum-entropy state of a qubit: measure any axis and get fifty-fifty. Alice’s particle carries no information about anything — not about Bob, not about the axis, not even about which of the four maximally entangled states the pair is in. The whole is as specific as a quantum state can be. The part is as featureless as a quantum state can be.

That is the green and pink of this series arriving somewhere new. What Alice can feel is her reduced state, and for a maximally entangled pair that is precisely nothing. The correlation is real, invariant, and perpendicular to every instrument she owns.

Figure 02

The more the whole has, the less the parts have

• entanglement entropy of |Ψ⟩ = cos θ |00⟩ + sin θ |11⟩
• the two angles at which the pair really is two things

05 Nothing happens at Bob’s end

Now the sentence that does more damage than any other here: measuring Alice’s particle does something to Bob’s. Every popular treatment implies it. It is false, and what makes it false is not a practical limitation but a theorem. Bob’s reduced state after Alice does anything at all to her half is:

Every probability of every outcome Bob can obtain is a function of ρ_B, and ρ_B did not move. Not slightly, and not in a way better technology would reveal. His statistics are identical in a world where Alice measured along x, a world where she measured along z, and a world where she went for a walk instead.

The assumptions are the whole content, so state them: systems compose by the tensor product, Alice’s operation acts as ℰ_A ⊗ I_B with no interaction term, her map is trace-preserving — we average over her outcomes rather than discarding some — and the Born rule holds. That third condition is where every faster-than-light proposal in the literature quietly goes wrong. Post-select on Alice’s results and Bob’s state appears to change; but selecting requires knowing, and knowing requires a message.

Note the boundary. This is Bob’s unconditioned state. Conditioned on Alice’s outcome his description certainly changes — that is steering, and it was Schrödinger’s own topic in the 1935 paper. He simply cannot condition on anything until an ordinary classical message reaches him at or below the speed of light.

A second reason to distrust the causal story comes from part five. The two measurements are spacelike separated, so their order is frame-dependent: there are good Lorentz frames in which hers is first and good ones in which his is. “Alice’s measurement collapsed Bob’s particle” names no frame-invariant fact, and neither does the reverse. Be honest about how far that reaches: it does not refute collapse-as-a-causal-process — you can posit a preferred slicing of spacetime and pay in structure no experiment can see, which is exactly what Bohmian mechanics does. It shows that relativity supplies nothing to underwrite the causal story. If you want it, you are buying it yourself.

The correlation is discovered, not transmitted. Nothing goes from Alice to Bob, and there is no frame-invariant fact about which of them went first.

06 Bell, stated exactly

If nothing travels and nothing happens at Bob’s end, why is anyone excited? Because the correlations, useless as they are for sending anything, cannot be reproduced by any story in which each particle carried its answers with it.

Alice picks one of two measurement directions, Bob picks one of two, each records +1 or −1, and they build a combination of the four correlation averages.

For a spin-½ singlet with the directions in a plane, quantum mechanics gives E(a,b) = −cos(a − b). Space the four settings θ apart and the combination becomes S(θ) = 3cos θ − cos 3θ — worth differentiating yourself, because the maximum sits at exactly 45° and its value is exactly 2√2.

Figure 03

The gap quantum mechanics opens, and the bigger one it declines

• the classical ceiling
• S(θ) = 3cos θ − cos 3θ, computed every 1.5°

Experiment settled the empirical question a decade ago. Three loophole-free tests landed in 2015 — electron spins 1.3 kilometres apart in Delft, and two photon experiments — closing the detection and locality loopholes together. In 2018 a hundred thousand volunteers generated the settings by hand for the BIG Bell Test, and another experiment drew them from the light of high-redshift quasars, pushing any conspiracy in the settings billions of years into the past. The 2022 Nobel went to Aspect, Clauser and Zeilinger “for experiments with entangled photons, establishing the violation of Bell inequalities and pioneering quantum information science.”

So what has been ruled out? Here the popular record is at its worst, and the answer is a list rather than a slogan. A Bell inequality follows from three assumptions, and violating it means at least one of them is false.

Figure 04

Four exits, not two

1. Deny locality Outcomes here do not depend on settings there. → Bohmian mechanics, collapse theories
   → Bell’s own choice
2. Deny single outcomes Each measurement has one definite result. → Everett
   → escapes without giving up locality
3. Deny measurement independence The settings are uncorrelated with the hidden state. → superdeterminism
   → a conspiracy in the common past
4. Deny it from the other end The same assumption, broken in the other temporal direction. → retrocausal models
   → the hidden state depends on future settings

Two consequences worth stating flatly. Bell did not refute determinism, and he did not refute hidden variables. Bohmian mechanics is fully deterministic, gives every particle a definite position at all times, reproduces the predictions of non-relativistic quantum mechanics, and survives Bell by being explicitly non-local — non-local in a way that, by section 05, cannot be used to send anything.

Bell spent years insisting on this and mostly failing to be heard. From “Bertlmann’s socks,” 1981: “It is important to note that to the limited degree to which determinism plays a role in the EPR argument, it is not assumed but inferred. What is held sacred is the principle of ‘local causality’.” In a note to the same paper he complains that commentators have “almost universally reported” his 1964 paper as beginning with deterministic hidden variables, when it begins with locality and derives them. In fairness, whether the formal derivation is as innocent of determinism as he insisted is a live dispute — Wiseman argues it is not, Norsen defends Bell’s reading. What nobody disputes is that the general result does not need determinism at all: the CHSH form drops the perfect-correlation step entirely, which is exactly why real experiments, which never achieve perfect correlations, can test it.

07 Entangled is not the same as non-local

Now the section that keeps this lesson honest. It is easy to slide from entangled to violates Bell to non-local as though they were one property with three names. They are not, and there is an explicit counterexample: the Werner state, a singlet mixed with pure noise in proportion p.

Figure 05

The window where a state is entangled and provably local

• Separable — a mixture of products, no entanglement at all
• Entangled, and a local hidden-variable model exists
• Entangled, no known local model, CHSH still not violated
• CHSH violated

• p = 1/3
• p = 1/2
• p = 1/√2
• p = 1

So the amber band exists: states entangled by the definition in section 03 that nevertheless admit an explicit local hidden-variable model. Entanglement and non-locality are different properties. Entanglement is a fact about the state’s algebraic form; Bell non-locality is a fact about which correlations it can produce. One implies the other only for pure states.

And there is a trap inside the trap. Failing to violate CHSH is far weaker than being local, because CHSH is one inequality among infinitely many and a state can slip past it while failing a cleverer one. That is why the number carrying the argument is 1/2 — Werner’s constructed model — and not 1/√2. Anyone who tells you the entangled-but-local window runs to 0.707 has swapped a proof for the absence of one particular refutation.

This is the move this series makes in every part, and I would rather make it against my own thesis than have a reader make it for me. A good picture, taken literally, becomes a bad belief. Entanglement-as-spooky-connection is a good picture. Taken literally it predicts that everything entangled is non-local, and that is false.

08 What actually fails

One more structural fact first, because it inverts what most people believe and takes two lines. If Alice and Bob share a pure state, then for any third system C the total state is ρ_ABC = ρ_AB ⊗ ρ_C. C is not merely unentangled with the pair; C is entirely uncorrelated with it, classical correlation included. That follows from purity alone, and the Coffman–Kundu–Wootters inequality of 2000 is its graded version: entanglement A spends on B is entanglement A cannot spend on C. Strong entanglement is exclusive. It is the least promiscuous relation in physics.

Now the metaphysics — and now I have to disappoint the version of this lesson I originally wanted to write.

The tempting claim is that for an entangled pair the parts are not there. It is vivid, it flatters the mathematics, and it is an overreach. The particles are there. You bought two, you detect two, each has a definite mass and charge and spin magnitude, and each gives a definite reading. The reduced density matrix is a perfectly good state; it is merely mixed.

Here is the claim that is exactly true.

Hand me the complete intrinsic state of Alice’s particle and of Bob’s, and tell me where each one is. I still cannot say which of the four maximally entangled states they are in, because all four give the same pair of reduced states and all four make different predictions. The information is not hiding in the parts. There is more in the whole than the parts and their arrangement contain.

That is a failure of supervenience, not of existence, and the distinction is the whole game. Collapse it and you have said something exciting and false. Keep it and you have said something modest and, as philosophy of physics goes, close to uncontroversial — Howard put it this way in 1985, Teller developed it as relational holism in 1986, Healey gave the canonical treatment in 1991.

Which brings the question this series exists to press. If structure is prior to objects — if mathematics is not a description laid over the world but the thing the world is — then entanglement looks like the best evidence anyone could ask for. All the information in the relations, none in the relata. Ontic structural realism with an equation attached. I think that reading is available. Three things stand against leaning on it too hard.

Get the attribution right. Ontic structural realism was named by James Ladyman in 1998 and developed with Steven French, but their primary quantum argument is about permutation invariance and the individuality of indistinguishable particles, not about entanglement. The entanglement-based argument belongs mainly to Michael Esfeld, from 2004 — and Esfeld and Lam’s version is deliberately moderate: it keeps the relata and claims only that the relations do not reduce to them.

The objection I find hardest to answer. To say a state is entangled you must first fix a decomposition ℋ = ℋ_A ⊗ ℋ_B. You must individuate the parts before the relation is even definable. Worse, the decomposition is not unique — a global unitary turns an entangled state into a product state, so entanglement is a property of a state together with a choice of parts. The choice is not arbitrary; it is fixed by which operations are locally implementable and by the locality structure of the Hamiltonian, which is Zanardi’s point about virtual subsystems. But that is what makes the objection bite. The parts are supplied by physics before entanglement can be defined at all. A relation that presupposes its relata is a poor advertisement for relations without relata.

And there is a rival that fits the mathematics at least as well. On Wallace and Timpson’s spacetime state realism, every region of spacetime carries its own density operator as an intrinsic property. The parts then do have states of their own; they simply fail to determine the whole. Which is the title of this lesson, arrived at from the opposite metaphysical direction, and it is the version I would defend if pressed.

So, the honest position. Quantum mechanics is compatible with, and suggestive of, the view that structure is prior to objects. It does not prove it. An object-oriented realist can accept every equation in this lesson and hold that the parts exist with intrinsic natures and that what failed was only the classical assumption that wholes are built from them by arrangement. Losing supervenience is a heavy enough blow to atomism without pretending we have also lost the atoms.

09 What it will not do

And now the temptation, which I want to name precisely rather than sneer at, because the impulse behind it is not silly.

The next step — from non-separability to influence, communion, or mind — does not go through, and four separate facts block it.

No-signalling. Entanglement produces correlation, never usable influence. Each party’s record is individually indistinguishable from a coin, and the correlation appears only when two records are brought together over an ordinary classical channel. Nothing happens at Bob’s end when Alice measures: a theorem about his density matrix, not a limit awaiting better instruments.

Monogamy. “Everything is entangled with everything” is the reverse of what section 08 proves. Strong entanglement leaves the rest of the universe entirely uncorrelated.

It has a supply chain. Entanglement does not spring up between distant systems that have never interacted. It is created by local interaction and carried apart, and even entanglement swapping — the trick that seems to make it appear between strangers — consumes pre-existing entanglement plus classical communication. A resource with logistics, not an ambient condition of the cosmos.

Fragility. Entanglement is destroyed by exactly the contact with a warm, wet, massive environment that a brain provides continuously. Tegmark’s estimate puts neural decoherence between 10⁻¹³ and 10⁻²⁰ seconds against cognitive timescales of 10⁻³ to 10⁻¹ — ten or more orders of magnitude adrift, and the published rebuttals do not close it.

The generous conclusion is also the true one. Entanglement refutes an atomism about states, and that is substantial — more substantial, I would argue, than the quantum indeterminacy that part seven found so much less generous than its enthusiasts hoped. But it says nothing about minds, and it supplies no connection anyone can feel, use, or send anything through.

You only feel your own reduced state. For a maximally entangled pair that is exactly nothing — and it is nothing in a way no instrument will ever improve upon.

10 The ledger

What is load-bearing

The tensor product and everything that follows from it by algebra: dimensions multiply, product states form a measure-zero submanifold, the Schmidt rank settles the question in one integer, the reduced state of a maximally entangled pair is exactly I/2, no-signalling is a theorem about trace-preserving maps, a pure ρ_AB forces ρ_ABC = ρ_AB ⊗ ρ_C, and the CHSH bounds are 2 and 2√2 with experiment exceeding the first. None of this depends on an interpretation, and none of it is in dispute.

What is convention

The sign in the singlet, and Alice and Bob as the two parties. Less innocently: the tensor factorisation itself — which degrees of freedom count as “the parts.” That is constrained by what is locally accessible rather than free, but it is settled before the word entanglement means anything, and section 08 is where it does damage.

Where the shorthand breaks

“Entangled” is not “non-local”: Werner states between p = 1/3 and p = 1/2 are entangled and have an explicit local model. “Measuring here affects there” is barred by a theorem and has no frame-invariant meaning. “Bell refuted determinism” is wrong — he refuted local hidden variables and said so repeatedly. And none of this requires complex numbers: the state (|00⟩ + |11⟩)/√2 has entirely real coefficients.

Where I would push back on myself

The strongest objection is not to any equation here but to the metaphysics I am tempted to hang on them. An object-oriented realist grants everything above and answers: the two particles exist, they are countable, each has intrinsic properties, each yields a definite outcome, and the reduced density matrix is a genuine state. What I have shown is that wholes are not built from parts by arrangement — which refutes a doctrine about composition, not the existence of the components. And they have a technical point I cannot dissolve: the tensor factorisation must be fixed before entanglement is definable, so the relation presupposes its relata rather than replacing them. I think non-separability is the more important discovery. I do not think it gets me to relations without relata, and I would be misleading you to imply otherwise.

11 Exercises

1. Count the room yourself Verify the two dimension counts in Figure 01 for N = 2 and N = 3. Then explain, in one paragraph and without using the word entanglement, why the classical Cartesian product and the quantum tensor product have to give different answers.
2. Trace out the partner Compute the reduced density matrix of one qubit for all four maximally entangled two-qubit states. You will get the same answer four times. Say precisely what that rules out about where the difference between those states is stored.
3. Find the assumption that fails State the three assumptions behind a Bell inequality as three separate propositions. For each of the four exits in Figure 04, say which proposition it denies and what the denial costs. Then say which cost you find least tolerable, and whether that is evidence about the world or about you.
4. Separate the two claims Write out “the parts are not there” and “the whole does not supervene on the parts” as two distinct theses. Construct a case that satisfies the second and refutes the first. What does that tell you about how much metaphysics a physics result can carry?
5. Argue the other side Take the object-oriented realist position from the ledger — the particles exist, they have intrinsic natures, and only classical composition has failed — and defend it against sections 04, 05 and 08 on its own terms. Then say what it costs you: on that view, what is the physical fact distinguishing the four maximally entangled states, and where does it live?