Patch project ยท 13 November 2026
The Patch and the Qubit
You are given an urn that contains a ball that is either white or black. You are given no other information about how the urn was prepared. What is the probability that the ball is black?
This probability cannot be expressed as a number. It is a new kind of mathematical entity that Ed Thorp suggested we call a “patch.” All the usual probability calculations can be done with patches. The patch is the simplest form of ignorance. All states of ignorance can be expressed using patches.
The question is: Can patches be reduced to qubits, can qubits be reduced to patches, or does quantum mechanics require something new?
Our answer, argued below: patches reduce to qubits; qubits do not reduce to patches; and the new thing quantum mechanics requires is small and can be stated exactly. It is not a new kind of ignorance. It is a different rule for the weights a patch carries. Along the way we ask what situation gives the qubit as the urn gives the patch: it is an urn that can be opened along any direction but holds only one bit.
1. The patch
Looking into the urn is a source of bits: write 1 if the ball is black and 0 if it is white. Look again and the bit is the same. Let x be the probability that the next bit a source yields is a 1. For a coin of known bias, x is a single number. For the urn it is not: the ball is black or it is white, so x is 1 or 0, and we are not told which. What we can write down is a density over x, and the source is a patch when that density is
a δ(x) + b δ(x − 1), a + b = 1.
All the weight sits at the two ends of the interval. Nothing about the source is random; the only uncertainty is ours. The patch is the whole expression, not either of its weights. Since nothing was said about how the urn was prepared, a and b are not given as numbers; they are carried through a calculation as symbols.
Calculations go through in terms of those symbols:
- Output. The sequence of bits is 000… with weight a or 111… with weight b. No other sequence occurs.
- Learning. One observation settles everything. After a 1 the density is δ(x − 1); after a 0 it is δ(x).
- Information. The first bit carries H(b) = −a log2 a − b log2 b bits. Every later bit carries none.
- Spread. The mean of x is b and its variance is ab. No density on the interval from 0 to 1 with the same mean has a larger variance [1].
- Betting. At even odds the Kelly bet [2] on the first bit is the fraction |b − a| of the bankroll, on the side with the larger weight. Every bet after that is a sure thing.
2. The patch and the coin
The opposite extreme is the density δ(x − b): a coin whose bias b is known exactly. Before the first bit the coin and a patch with the same b lead to the same bet. After it they could not be more different. The coin teaches nothing however long it is watched, and the patch teaches everything at once.
| Density of x | Weight on a 1 first | Weight on a 1 after seeing a 1 | Variance of x |
|---|---|---|---|
| Coin, δ(x − ½) | ½ | ½ | 0 |
| Uniform on 0 to 1 | ½ | ⅔ | 1/12 |
| Patch, a = b = ½ | ½ | 1 | ¼ |
The coin is chance with nothing left to learn. The patch is ignorance with no chance in it.
3. All classical ignorance is made of patches
A patch is ignorance about one question with two answers, where the answer is already fixed. Larger states of ignorance are built from it.
More than two answers. Let the ball be white, black or red. Ask “is it red?”; that is a patch. If the answer is no, ask “is it black?”; that is a second patch. Any state of ignorance over a finite list of fixed alternatives is a short tree of patches of this kind, and the entropy adds up along the tree.
One way only. However the tree is drawn, the result is the same list of weights, one for each fixed alternative. A classical state of ignorance is a mixture of decided cases, and it is that mixture in exactly one way. This is the precise sense in which every classical state of ignorance can be expressed using patches. It is also the property that quantum mechanics will turn out not to have.
Sources that repeat. A sequence of bits whose weights do not depend on the order of the bits can always be written as a coin of unknown bias, with some density over x (de Finetti's theorem [3]). Among such densities the coin and the patch are the two ends of the range.
4. The urn that gives the qubit
The urn of the opening lets us derive the patch. The situation that lets us derive the qubit is an urn that can be opened along any direction but holds only one bit.
The physical example is the experiment of Stern and Gerlach [8]. A particle of spin ½ passes through a magnet set along a direction of our choosing and is deflected up or down. Sent through a second magnet set the same way, it repeats its answer. That is the urn of the opening, except that there are many ways to open it. Three conditions describe it:
- Every opening is a patch. Whichever direction is chosen, the answer is one of two, and it repeats.
- One bit only. The answer along at most one direction (and its opposite) can be settled at a time. Settling a new one unsettles the old.
- No preferred direction. All openings are equivalent, and one can be turned smoothly into another.
The qubit follows in two steps. First, a state of complete knowledge is one in which some direction's answer is settled, so the states of complete knowledge are the directions themselves and form a sphere. States of ignorance are mixtures of these and fill the ball inside it. A state is a vector r of length at most 1. Second, weights combine linearly when states are mixed, as they do for patches. The weight on “yes along n” must then be a linear function of r that is 1 at r = n, 0 at r = −n, and unchanged by turning about n. Only one function does this:
weight on “yes along n” = (1 + r·n)/2.
This is the quantum rule for a qubit, here derived and not assumed. The ball is the Bloch ball, and the usual α|0〉 + β|1〉 is another way of naming a point on its surface. This is a sketch; the complete derivations are in [9–14].
The patch is the same construction with a single opening.
| Ways to open the urn | States | Object |
|---|---|---|
| One | A line segment | Patch |
| Around a circle | A disc | Qubit with real amplitudes only |
| Over a sphere | A ball | Qubit |
The three conditions do not say why nature uses the sphere and not the circle or a sphere of more dimensions. The published derivations settle it by a further requirement: that two such urns can interact reversibly and become entangled [10, 12, 13].
The urn of the opening, prepared in an unknown way, gave a patch whose weights are unknown. The same is true here. A particle prepared in an unknown way is a qubit whose r is unknown: three unknown numbers where the patch has one [15].
5. Can patches be reduced to qubits? Yes
In the usual notation, a qubit in the state α|0〉 + β|1〉, with |α|2 + |β|2 = 1, is measured in the 0/1 basis. The result is 1 with probability |β|2. The measurement leaves the qubit in the state it reported, so every later measurement in the same basis gives the same result. The record is all 0s or all 1s: a patch with a = |α|2 and b = |β|2.
| Patch | Qubit, measured repeatedly in one basis |
|---|---|
| Weights a and b | Squared amplitudes |α|2 and |β|2 |
| Seeing the first bit moves the density to δ(x) or δ(x − 1) | The first measurement projects the state onto |0〉 or |1〉 |
| The first bit carries H(b) bits, later bits none | The first measurement yields H(|β|2) bits, repeats none |
| The coin δ(x − b) | Many qubits prepared the same way, each measured once |
The patch is the qubit with its phase removed. The patch keeps |α|2 and |β|2 and drops the relative phase of α and β. As a density matrix it is the diagonal part, a|0〉〈0| + b|1〉〈1|. Mixing patches, combining them and conditioning on an observation are all reproduced by diagonal density matrices. Ordinary probability sits inside quantum mechanics as the special case in which everything commutes.
Measurement makes the patch. To measure the qubit is to copy its bit onto something that keeps a record. If the record is a qubit starting in |0〉, the copy turns α|0〉 + β|1〉 into the entangled pair α|00〉 + β|11〉, and each further measurement adds a record: α|000…〉 + β|111…〉. The patch's rule, all 0s or all 1s, is the perfect correlation of this state. Any one record taken alone is the diagonal state above. The phase has not been destroyed; it has moved into the joint state, where no single record can see it. The entanglement between the qubit and its record, measured as entropy, is H(b), the same number of bits the first look at the patch delivers.
One thing the qubit does not supply. The opening says the probability for the urn cannot be expressed as a number. A qubit does not explain that. Once a state is assigned to it, quantum mechanics returns a number for every measurement. An urn prepared in an unknown way corresponds to a qubit in an unknown state, and that is ignorance of the patch kind placed on top of the qubit, not derived from it [15]. So the reduction holds for patches whose weights are given, and for patches with symbolic weights it holds symbol for symbol.
6. Can qubits be reduced to patches? No
To reduce a qubit to patches would be to say that every result it can give is a bit already decided and merely unknown to us. How far that can be carried depends on the system.
A qubit holds a patch for every direction. A qubit can be measured along any axis. If its state has Bloch vector r and the measurement is along the unit vector n, repeated measurement gives a patch whose weight on the result “along n” is (1 + r·n)/2. These patches cannot all be settled. Measure along Z and the Z patch is known. Then measure along X: the result is 0 or 1 with weight ½ each, and the Z patch has been reset to a = b = ½. Alternate Z and X and the one qubit becomes a fair coin. A classical patch, once seen, stays seen.
One qubit: the reduction can be forced through. A model exists in which a hidden variable fixes the result for every axis in advance and reproduces all these statistics (Bell, 1966 [4]). Its price is that each measurement must scramble the hidden variable, so the decided bits are never the ones found twice.
One qutrit: it fails. A qutrit has three basis states. Each direction v in its three-dimensional state space gives a bit patch, “are you along v?”, with weight |〈v|ψ〉|2. Unlike the qubit's, some of these patches are compatible: those for perpendicular directions can be settled together, and of any perpendicular three exactly one answers 1. For a particle of spin 1 they are the squares of the spin along x, y and z, which always come out as 1, 1 and 0 in some order. Now suppose every such patch were already decided, each direction carrying its 0 or 1. The Kochen–Specker theorem [5] shows that no assignment satisfies the one-in-three rule; a few dozen well-chosen directions force a contradiction [6, 19]. The bit along v can be fixed in advance only if it is allowed to depend on which other two directions are asked about with it.
An entangled pair: it fails again. Two people share the singlet state, (|01〉 − |10〉)/√2. Whatever axis they choose, if they choose the same one their results are opposite: a shared patch for every direction at once. A shared patch for one direction is easy to make classically, by writing a hidden bit on two cards. Bell's theorem (1964 [7]) says that no set of bits decided in advance, one for each direction, reproduces what the pair does when the two people choose different axes, unless settling a patch in one place can alter a patch in the other.
So qubits reduce to patches only if the patches are allowed to disturb one another: by being measured, by being asked about in different company, or at a distance. Patches as defined in section 1, fixed answers that are simply not yet known, are not enough.
7. What quantum mechanics adds
The mixture is no longer unique. Section 3 said a classical state of ignorance is a mixture of decided cases in exactly one way. Quantum states are not. A qubit that is a fair patch along Z, ½|0〉〈0| + ½|1〉〈1|, is the very same state as a fair patch along X; no measurement tells them apart. There is no fact about which decided cases the ignorance is ignorance of.
Complete knowledge leaves patches unsettled. The state |0〉 is as much as can be known about a qubit. Its Z patch is settled. Its X patch has weights ½ and ½, and no further information about the qubit will change that. Classically, complete knowledge settles every patch. In quantum mechanics it settles one family and leaves the rest open.
Weights that add before they are squared. The state (|0〉 + |1〉)/√2 and the fair patch give the same weights in a Z measurement. Measured along X, the patch gives ½ and ½, and the superposition gives one result with certainty. The difference is that the patch adds weights and the superposition adds amplitudes, and amplitudes can cancel.
The same thing said with signed weights. The form of the patch can be kept. A qubit's state can be written as weights on four decided cases, a value for Z together with a value for X, in such a way that every measurement's statistics come out as ordinary sums of the weights [16]. The cost is that for some states one of the weights is negative. For qutrits, the states that need a negative weight are exactly the ones that show the Kochen–Specker kind of dependence under a standard set of measurements [17]. On this view quantum mechanics is the theory of patches whose weights need not lie between 0 and 1.
The weights are not free. Require only that every direction of a qutrit has a weight between 0 and 1 and that the weights of any perpendicular three add to 1. Gleason's theorem [18] says the weights must then be the quantum ones. Once patches for different directions are tied together by that one rule, the rest of the quantum rule for probabilities follows.
8. The answer
- Patches reduce to qubits. A patch is a qubit seen in one basis, with its phase removed, and all of ordinary probability is the commuting part of quantum mechanics. What the qubit does not supply is the absence of a number: an urn of unknown preparation is a qubit in an unknown state.
- Qubits do not reduce to patches. One qubit can be forced into the form, a qutrit cannot without letting a patch depend on its company, and an entangled pair cannot without letting it depend on events elsewhere.
- Something new is required, and it is small. Quantum mechanics keeps the patch and changes the arithmetic of its weights: amplitudes that add before squaring, or equivalently real weights that may go negative. The consequences are that ignorance no longer splits into decided cases in a unique way, and that the fullest possible knowledge still leaves patches open.
Notes. The urn problem, the statements about the patch in the opening, Ed Thorp's name for it and the question were given to us as they stand there. The answer is our own analysis and is not attributed to him; in particular, treating the weights a and b as symbols is our reading of “cannot be expressed as a number.” The derivation of the qubit in section 4 is our sketch of results in the published reconstructions of quantum theory [9–14]. We know of no published source for the patch itself.
Bibliography
- R. Bhatia and C. Davis, “A better bound on the variance,” American Mathematical Monthly 107 (2000) 353–357. doi:10.1080/00029890.2000.12005203. The largest variance of a quantity confined to an interval.
- J. L. Kelly, “A new interpretation of information rate,” Bell System Technical Journal 35 (1956) 917–926. doi:10.1002/j.1538-7305.1956.tb03809.x. The Kelly bet.
- B. de Finetti, “La prévision: ses lois logiques, ses sources subjectives,” Annales de l'Institut Henri Poincaré 7 (1937) 1–68. numdam.org/item/AIHP_1937__7_1_1_0. Order-independent sequences as coins of unknown bias.
- J. S. Bell, “On the problem of hidden variables in quantum mechanics,” Reviews of Modern Physics 38 (1966) 447–452. doi:10.1103/RevModPhys.38.447. The hidden-variable model of a single qubit.
- S. Kochen and E. P. Specker, “The problem of hidden variables in quantum mechanics,” Journal of Mathematics and Mechanics 17 (1967) 59–87. doi:10.1512/iumj.1968.17.17004. The original proof, with 117 directions.
- A. Peres, “Two simple proofs of the Kochen–Specker theorem,” Journal of Physics A 24 (1991) L175–L178. doi:10.1088/0305-4470/24/4/003. A proof with 33 directions.
- J. S. Bell, “On the Einstein Podolsky Rosen paradox,” Physics 1 (1964) 195–200. doi:10.1103/PhysicsPhysiqueFizika.1.195. Bell's theorem for entangled pairs.
- W. Gerlach and O. Stern, “Der experimentelle Nachweis der Richtungsquantelung im Magnetfeld,” Zeitschrift für Physik 9 (1922) 349–352. doi:10.1007/BF01326983. The experiment.
- A. Zeilinger, “A foundational principle for quantum mechanics,” Foundations of Physics 29 (1999) 631–643. doi:10.1023/A:1018820410908. An elementary system carries one bit.
- B. Dakić and Č. Brukner, “Quantum theory and beyond: is entanglement special?” (2009). arXiv:0911.0695. The ball of states, and why it has three dimensions.
- L. Hardy, “Quantum theory from five reasonable axioms” (2001). arXiv:quant-ph/0101012. The first of the modern reconstructions.
- Ll. Masanes and M. P. Müller, “A derivation of quantum theory from physical requirements,” New Journal of Physics 13 (2011) 063001. arXiv:1004.1483.
- Ll. Masanes, M. P. Müller, R. Augusiak and D. Pérez-García, “Existence of an information unit as a postulate of quantum theory,” Proceedings of the National Academy of Sciences 110 (2013) 16373. arXiv:1208.0493.
- M. P. Müller and Ll. Masanes, “Three-dimensionality of space and the quantum bit: an information-theoretic approach,” New Journal of Physics 15 (2013) 053040. arXiv:1206.0630.
- C. M. Caves, C. A. Fuchs and R. Schack, “Unknown quantum states: the quantum de Finetti representation,” Journal of Mathematical Physics 43 (2002) 4537. arXiv:quant-ph/0104088. What it means for a quantum state to be unknown.
- W. K. Wootters, “A Wigner-function formulation of finite-state quantum mechanics,” Annals of Physics 176 (1987) 1–21. doi:10.1016/0003-4916(87)90176-X. Signed weights on decided cases.
- M. Howard, J. Wallman, V. Veitch and J. Emerson, “Contextuality supplies the ‘magic’ for quantum computation,” Nature 510 (2014) 351–355. arXiv:1401.4174. Negative weights and Kochen–Specker dependence coincide for qutrits.
- A. M. Gleason, “Measures on the closed subspaces of a Hilbert space,” Journal of Mathematics and Mechanics 6 (1957) 885–893. doi:10.1512/iumj.1957.6.56050.
- A. Peres, Quantum Theory: Concepts and Methods, Kluwer, 1993. doi:10.1007/0-306-47120-5. Reports the 31-direction proof of Conway and Kochen.