---
title: Composition Impossibility Theorem
url: https://www.emergentmind.com/topics/composition-impossibility-theorem
type: topic
---

# Composition Impossibility Theorem

Searching arXiv for the cited papers to ground the article in the published sources.
{"query":"id:1806.10374 OR id:1604.01386 OR id:1306.5805 OR id:2106.11200 OR id:1908.00972","max_results":10}
I’m checking the exact arXiv records and metadata for the five papers referenced in the source material.
“Composition impossibility theorem” is not a single theorem with one formal statement, but a recurring no-go pattern in which a seemingly natural rule for building a composite object from well-defined constituents fails under basic assumptions. In the cited literature, the phrase and closely related formulations occur in quantum process theory, deterministic hidden-variables models, algebra of binary relations, relativistic quantum cryptography, and the theory of radicals and polynomial solvability. In each setting, the obstruction concerns a specific compositional principle: parallel tensoring of process matrices, subsystem-to-composite ontic composition, finite relational representation under composition and intersection, composable realization of oblivious transfer from bare communication, or finite nesting of radicals to reconstruct generic roots [1806.10374] [1306.5805] [1604.01386] [2106.11200] [1908.00972].

## 1. Recurring structure of composition impossibility

A common structure links these otherwise disparate results. One begins with local or single-instance objects that are individually valid: process matrices, ontic states, binary relations, cryptographic resources, or analytic/radical expressions. One then posits a rule that should build a valid global object while preserving basic consistency with known special cases. The theorem shows that the rule is either unique and invalid outside a restricted regime, or directly inconsistent with the ambient formalism.

| Domain | Objects being composed | Obstruction |
|---|---|---|
| Quantum processes | Process matrices $W,W'$ | The only bilinear, order-consistent candidate is $W \otimes W'$, but it fails the process constraints for indefinite causal order |
| Hidden-variables models | Ontic states $\lambda_1,\lambda_2$ | $PI_{c,tr}$ leads to violation of $\sum_k pr(M=k \mid \lambda)=1$ |
| Relation algebras | Composition $;$ and intersection $\cdot$ | A finite algebra is representable only on an infinite base |
| Relativistic cryptography | Protocol/resource composition for OT | No composable OT from bare communication for $\epsilon<1/12$ |
| Abel–Ruffini setting | Finite compositions of analytic functions and radicals | Such compositions have solvable monodromy, but generic degree-$n$ roots have monodromy $S_n$ for $n \ge 5$ |

This shared pattern suggests that “composition impossibility” typically does not mean that composite systems are meaningless or unusable. It means that a preferred composition rule, chosen for its apparent naturality or consistency with simpler cases, is incompatible with the formal constraints of the theory under consideration.

## 2. Quantum processes and the failure of a natural tensor product

In the process-matrix formalism, a quantum process matrix $W \in L(H_A \otimes H_B)$ is the most general object assigning joint probabilities to local quantum operations without assuming a background causal order. For two parties $A$ and $B$, with input spaces $A_I,B_I$ and output spaces $A_O,B_O$, the spaces are $H_A = H_{A_I} \otimes H_{A_O}$ and $H_B = H_{B_I} \otimes H_{B_O}$. A valid process satisfies positivity $W \succeq 0$, trace normalization $\mathrm{Tr}\,W = d_{A_O}d_{B_O}$, and the “no-loop” linear constraint $L_V(W)=W$, where $L_V$ projects onto the linear subspace defined by the paper’s constraints. Local operations are represented by Choi matrices $M_{a|x}$, and the generalized Born rule is
$$
p(ab|xy)=\mathrm{Tr}\!\big[(M^{A}_{a|x}\otimes M^{B}_{b|y})\,W^T\big].
$$
This reduces to the standard state- and channel-based probability rules in fixed-order cases [1806.10374].

The central question is whether there exists a binary map $\mu:(W,W')\mapsto \mu(W,W')$ giving a parallel composition of two processes for the same parties. The desired properties are: validity, meaning $\mu(W,W')$ is itself a valid bipartite process; consistency, meaning that if $W$ and $W'$ have the same definite causal order then $\mu(W,W')=W\otimes W'$; and convex linearity, meaning that composition commutes with mixing. Under the relaxed assumptions of positivity only, consistency on definite-order processes, and real-linearity in each argument, Theorem 1 shows that there is no bilinear map other than the ordinary tensor product:
$$
\mu(W,W')=W\otimes W'.
$$
The proof proceeds by extending convex linearity to a real-linear map on the full span, proving Hermiticity and operator-norm bounds, and then using PTI operators (“products of traceless or identity”) with eigenvalues in $\{-1,0,1\}$ to force tensor-product action on a generating basis [1806.10374].

The decisive obstruction appears in Theorem 2. Since the unique linear and order-consistent candidate is the tensor product, any valid rule must coincide with $W\otimes W'$. But for indefinite causal order this tensor product fails the “no-loop” constraint. The paper gives the example
$$
W=\tfrac12\bigl(W^{A\preceq B}+W^{B\preceq A}\bigr),
$$
for which $W\otimes W$ violates $L_V$ and therefore does not yield normalized probabilities; this is explicitly identified with the “grandfather paradox” causal loop. The consequence is that general quantum processes do not admit a bilinear, positivity-preserving, order-consistent parallel composition. The paper therefore concludes that a Shannon theory of general quantum processes cannot possess a natural rule for composing resources, in contrast with states and fixed-order channels, where tensor products are well defined [1806.10374].

## 3. Hidden-variable composition for product states

A distinct composition impossibility appears in deterministic hidden-variables theories. In the ontological framework, a pure state $|\psi\rangle$ is associated with a measurable ontic-state space $\Lambda_\psi \neq \varnothing$ and a probability density $p_\psi(\lambda)>0$ on $\Lambda_\psi$, with $\int_{\Lambda_\psi} p_\psi(\lambda)\,d\lambda = 1$. For an observable $M$ with spectrum $S(M)$, there is a response function
$$
pr(M=k \mid \lambda)=\Pr[\text{outcome}=k\mid \text{complete state }\lambda],
$$
subject to the no-“null-shot” condition
$$
\sum_{k\in S(M)} pr(M=k\mid \lambda)=1.
$$
The Born rule is recovered on average, and in the deterministic case one has $pr(M=k\mid \lambda)\in\{0,1\}$, so $M(\lambda)$ is well defined [1306.5805].

The composition principle at issue is preparation-independence in its compositional form. The weaker rule used in the theorem is $PI_{c,tr}$: if $\lambda_1$ tracks $|\psi_1\rangle$ on $S_1$ and $\lambda_2$ tracks $|\psi_2\rangle$ on $S_2$, then $(\lambda_1,\lambda_2)$ tracks $|\psi_1\rangle \otimes |\psi_2\rangle$ on $S_1 \otimes S_2$. Tracking means that for every $M$ and eigenvalue $k$,
$$
\Pr(M=k\mid |\phi\rangle)=0 \Longrightarrow pr(M=k\mid \lambda)=0.
$$
Together with determinism and the weak projector assumption $A$—if $\lambda\in\Lambda_\psi$ and $P_k(\lambda)=0$ then $M(\lambda)\neq k$—this suffices for the no-go argument [1306.5805].

The theorem states that any deterministic hidden-variables model on a two-dimensional Hilbert space $\mathcal H_0$ that reproduces quantum statistics on $\mathcal H_0$, satisfies assumption $A$, and satisfies $PI_{c,tr}$ cannot be extended to cover tensor-product states in $\mathcal H_0\otimes \mathcal H_0$ without contradicting $\sum_k pr(M=k\mid \lambda)=1$. The proof has three steps. First, for any distinct nonorthogonal states $|\psi\rangle,|\phi\rangle \in \mathcal H_0$, each $\Lambda_\psi$ contains a subset of measure $|\langle \psi \mid \phi\rangle|^2$ whose members also track $|\phi\rangle$. Second, choosing qubit states $|1\rangle,|2\rangle$ with $|\langle 1\mid 2\rangle|=2^{-1/2}$ yields a nonempty set of ontic states that track both single-qubit states; by $PI_{c,tr}$, $(\lambda,\lambda)$ then tracks all four product states $|x\rangle\otimes |y\rangle$, $x,y=1,2$. Third, the Pusey–Barrett–Rudolph measurement provides an orthonormal basis $\{|\xi_{xy}\rangle\}$ with $\langle \xi_{xy}\mid x,y\rangle=0$ for each pair $(x,y)$, so tracking forces zero response probability for all four outcomes, violating normalization [1306.5805].

The significance of this theorem is sharply delimited in the source. It does not target locality in the Bell sense or noncontextuality in the Bell–Kochen–Specker sense. Rather, it targets the classical-looking assumption that subsystem “real states” can be combined into a complete state for a nonentangled product system. The paper’s stated implication is that one must reject or modify $PI_{c,tr}$, for example by adding contextual or relational data that arise only at the composite level [1306.5805].

## 4. Composition and intersection in relation algebras

In algebraic logic, the composition impossibility theorem takes the form of a failure of the finite representation property (FRP). For a nonempty set $U$, an algebra of binary relations over $U$ is a set $A\subseteq \wp(U\times U)$ equipped with composition
$$
a;b=\{(x,y)\in U\times U:\exists z\in U\,[(x,z)\in a \wedge (z,y)\in b]\}
$$
and intersection
$$
a\cdot b=\{(x,y):(x,y)\in a \wedge (x,y)\in b\},
$$
and closed under both operations. If $K$ denotes the class of all abstract algebras of similarity type $(;,\cdot)$ that arise concretely in this way, then $K$ has the FRP if every finite representable algebra also has a representation over a finite base set. Maddux’s theorem states that $K$ does not have the finite representation property [1604.01386].

The counterexample is a finite three-element algebra
$$
A=(\{z,e,r\};,\cdot)
$$
represented over the infinite set $U=\mathbb Q$ by $z=\varnothing$, $e=\mathrm{Id}_U$, and $r=\{(x,y):x,y\in\mathbb Q \text{ and } x<y\}$. This algebra is closed under composition and intersection, with key equations
$$
r;e=r=e;r,\qquad r;r=r,\qquad z;r=z=r;z,\qquad r\cdot e=z.
$$
The proof shows that any abstract algebra satisfying these equations must be represented on an infinite set. Thus a finite abstract algebra exists that is representable over an infinite base but over no finite base [1604.01386].

The argument proceeds through a sequence of lemmas. First, one derives $Id_U\cdot r \subseteq z$. Next, one defines a diversity relation $D\subseteq U\times U$ as a relation disjoint from $Id_U$, and calls it dense if every $(u,v)\in D$ admits $w\in U$ with $(u,w),(w,v)\in D$. Then $r\cdot z$ is shown to be a nonempty dense diversity relation. From this, Proposition 4.4 constructs an infinite chain of distinct points $x_0,x_1,x_2,\dots,y\in U$ such that for all $i<j$,
$$
(x_i,x_j)\in r\cdot z \quad\text{and}\quad (x_i,y)\in r\cdot z.
$$
Hence $U$ must be infinite [1604.01386].

The importance of the example lies in its minimality and in the role of the two operations. The source emphasizes that the example uses the smallest nontrivial set of relations $\{z,e,r\}$ and only composition and intersection. It further states that if one omits intersection or replaces it by union, many finite representation theorems reappear. The theorem therefore isolates the simultaneous presence of $;$ and $\cdot$ as already sufficient to force infinite representability in some finite algebras [1604.01386].

## 5. Composable oblivious transfer in relativistic quantum cryptography

In relativistic quantum cryptography, the relevant impossibility concerns composable construction of oblivious transfer (OT). The formal setting is Abstract Cryptography implemented with Causal Boxes. A cryptographic resource $R$ is an abstract box with interfaces; converters $\alpha,\beta$ attached to the interfaces form $\alpha R \beta$; and distinguishers $D$ induce a distinguishing advantage
$$
d^{\mathcal D}(R,S):=\sup_{D\in\mathcal D}\bigl|\Pr[D[R]=1]-\Pr[D[S]=1]\bigr|.
$$
A protocol $\Pi=(\Pi_A,\Pi_B)$ $\epsilon$-constructs a target primitive from a setup if it satisfies correctness and simulation-based security against dishonest Alice and dishonest Bob. Causal boxes model quantum messages tagged by space–time points, with outputs at $P=(\vec x,t)$ depending only on inputs in the causal past of $P$; composition of causal boxes remains causal [2106.11200].

The paper formalizes several OT variants: Rabin OT, $p$-Rabin OT, $1$-out-of-$2$ OT, and Randomized OT (ROT). It proves equivalences among them: ROT $\to$ OT and OT $\to$ ROT are perfect; OT $\to OT^R$ is perfect; and $OT^R \to OT$ uses Crépeau’s reduction with $O(k)$ parallel Rabin-OT and distinguishing advantage at most $e^{-\Omega(k)}$. The main impossibility theorem is Theorem 3.1: for any $\epsilon<1/12$, there is no protocol using only relativistic quantum communication, or even general non-signalling communication, that composably constructs ROT. Corollary 3.2 transfers the same bound to $1$-out-of-$2$ OT, and Theorem 3.3 gives the $p$-Rabin OT bound $\epsilon < \tfrac12 \cdot p(1-p)\cdot (1/3)$, equivalently $\epsilon < p(1-p)/6$. For string-OT of length $s$, the bound sharpens to $\epsilon < \tfrac12(1-2^{-s})$ [2106.11200].

The proof is a simulation-based distinguishing argument. Assuming a protocol $\Pi$ $\epsilon$-constructs ROT, one obtains a composed resource $ROT_B\,\sigma_B\,\sigma_A\,ROT_A$ within $3\epsilon$ of the ideal ROT by the triangle inequality. A distinguisher then performs a man-in-the-middle attack: it chooses $b'\leftarrow\{0,1\}$, feeds it to the composed resource, receives $s'_{b'}$, and checks whether it equals the original $s_{b'}$. In the ideal ROT resource equality always holds, while in the simulated construction the simulator must guess $s'_{b'}$ with probability at least $1/4$ on the branch $b\neq b'$, because Minkowski causality places the required output outside the light-cone of the value available to the simulator. This yields distinguishing advantage at least $1/4$, so $3\epsilon \ge 1/4$ and therefore $\epsilon \ge 1/12$ [2106.11200].

The paper extends the impossibility to secure two-party computation via an AND-function primitive, and from there to multi-party computation. It also records a mutual construction between oblivious transfer and bit commitment: bit commitment $\to$ OT using $2n$ ideal commitments with distinguishing advantage $\le e^{-\Omega(n)}$, and OT $\to$ bit commitment using $k$ instances of $1$-out-of-$2$ OT with distinguishing advantage $\le 2^{-k}$. The stated interpretation is that no purely relativistic and quantum protocol can yield a composable OT or general MPC from bare communication alone; extra setup assumptions are required [2106.11200].

## 6. Radical composition and Abel–Ruffini

A classical version of composition impossibility appears in the Abel–Ruffini theorem. In the formulation given in the cited source, one seeks an algebraic expression
$$
E(a_0,\dots,a_n)=g\bigl(\sqrt[n_1]{f_1(a)},\sqrt[n_2]{f_2(a)},\dots,\sqrt[n_k]{f_k(a)}\bigr),
$$
where each $f_j:\mathbb C^{n+1}\to\mathbb C$ is analytic in the coefficients, $\sqrt[n_j]{\cdot}$ denotes an $n_j$th-root operation, and $g$ is analytic or built from field operations. The expression is intended to reconstruct the roots of a general polynomial from its coefficients. The precise impossibility statement is that for $n\ge 5$ there is no such finite composition of analytic functions and radical extractions that yields all roots of the general polynomial as single-valued functions of the coefficients [1908.00972].

The proof strategy is topological and group-theoretic. Let $Poly_n$ be coefficient space with the discriminant locus removed. A closed loop $\gamma$ in $Poly_n$ analytically continues each root function $x_j(a)$ to some $x_{\sigma(j)}(a)$, defining a monodromy representation
$$
\rho:\pi_1(Poly_n,a)\to S_n.
$$
For the general degree-$n$ polynomial, the image is the full symmetric group $S_n$. By contrast, a single $d$th-root extraction has cyclic monodromy of order $d$, and compositions built by field operations and radicals remain within a solvable extension tower. The decisive mismatch is therefore that any finite radical composition has solvable monodromy, whereas $S_n$ is non-solvable for $n\ge 5$ [1908.00972].

The source illustrates the contrast with low-degree formulas. The quadratic formula uses one square root, and its monodromy is a transposition of order $2$. Cardano’s cubic formula uses one square root and two cube roots, and its monodromy embeds into a solvable extension of cyclic groups of orders $2$ and $3$. For quintics, the source gives the family $x^5+a\,x+1=0$ with $a(t)=Re^{2\pi i t}$ for small $R>0$, along which the five roots undergo a $5$-cycle. This concretizes the impossibility: finite nesting of radicals cannot keep pace with the full $S_5$ monodromy [1908.00972].

The visualization paper makes this obstruction geometric. Using JavaScript and p5.js, it tracks loops in coefficient space whose analytic continuation returns radical-based expressions to their initial values while permuting the roots nontrivially. In the stated interpretation, the visualization does not replace the proof; it renders the monodromy obstruction explicit by showing the mismatch between single-valued radical compositions and the actual permutation behavior of generic roots [1908.00972].

## 7. Comparative interpretation and limits of the no-go claims

Taken together, these theorems identify several distinct senses in which composition can fail. In quantum process theory, the failure concerns a bilinear, positivity-preserving, order-consistent parallel composition for indefinite causal order; the tensor product remains valid on definite-order processes but not in general [1806.10374]. In deterministic hidden-variables theories, the contradiction is tied to $PI_{c,tr}$ and shows that subsystem ontic states do not straightforwardly determine the ontic description of a product system, even when the quantum state is nonentangled [1306.5805]. In algebraic logic, the obstruction is representational: closure under composition and intersection can force any representation of a finite algebra onto an infinite base [1604.01386]. In relativistic cryptography, the impossibility is composable and setup-sensitive: OT and MPC cannot be constructed from bare communication under Abstract Cryptography security, even with relativistic causality and quantum communication [2106.11200]. In Abel–Ruffini, the obstruction is not to solving particular equations, but to any single-valued finite composition of analytic functions and radicals that reconstructs roots of the generic polynomial of degree at least five [1908.00972].

A recurrent misconception is to read these theorems as blanket denials of composite structure. The sources support a narrower reading. The quantum-process result excludes a natural resource-composition rule but leaves open one-copy or single-shot information-theoretic approaches [1806.10374]. The hidden-variables result excludes a classical-looking compositional rule but explicitly points toward additional contextual or relational variables at the composite level [1306.5805]. The cryptographic result excludes construction from bare communication, not from stronger setup assumptions such as shared resources, relativistic multiple agents, bounded-storage, or trusted hardware [2106.11200]. The algebraic and Abel–Ruffini results likewise isolate precise formal constraints—finite representation over finite bases in one case, radical composition in the other—rather than denying the broader mathematical theories in which they arise [1604.01386] [1908.00972].

In that sense, “composition impossibility theorem” functions less as the name of one doctrine than as a recurring mathematical schema. The theorems differ in ontology, proof technique, and domain of application, but they converge on a single methodological lesson: a composition rule that is natural in simple or classical regimes can fail decisively when the global structure is governed by causal constraints, contextuality, representational closure, composable security, or monodromy.

Source: https://www.emergentmind.com/topics/composition-impossibility-theorem