Ordinary quantum circuits assume a background causal order: gate A acts before gate B, or B before A, or separate branches are classically selected. Quantum causal frameworks ask what mathematical structure remains if the order connecting local quantum operations is itself allowed to be nonclassical.
The process-matrix formalism separates local laboratory operations from the global object connecting them. Each laboratory is described by a completely positive map. The surrounding process assigns probabilities to combinations of those local maps. Some valid processes are compatible with one fixed order; some are classical mixtures of alternative orders; some are causally nonseparable and cannot be written as such a mixture.
Three ideas must be kept distinct. Causal nonseparability is an operator-level property of a process. Violation of a causal inequality is a device-independent statistical property showing that observed correlations cannot arise from any definite causal order under the inequality’s assumptions. The quantum switch is a physically important causally nonseparable higher-order process, but causal nonseparability alone does not imply causal-inequality violation.
Local quantum operations → Choi operators → global process matrix → probability rule → test causal separability → design witnesses or causal games → distinguish coherent control of order from classical random order.
1. Fixed causal order in an ordinary circuit
Suppose laboratory A receives a system, applies an operation, and passes its output to laboratory B.
The order is
A ≺ B.
Information can flow from A’s choice to B’s statistics, but not from B’s later choice back to A’s earlier observed outcome.
A different experiment may implement B≺A. A classical controller may randomly choose between the two orders. That random mixture still has a definite order in each run.
2. Local laboratories as quantum instruments
At laboratory A, an experimental choice x selects a quantum instrument {𝓜Aa|x} indexed by outcome a.
Each 𝓜Aa|x is completely positive and trace-nonincreasing. Summing over a gives a trace-preserving channel for each x.
Laboratory B is described similarly by {𝓜Bb|y}.
The global causal object should determine
p(a,b|x,y)
for arbitrary valid local instruments.
3. Choi–Jamiołkowski representation
A linear map 𝓜:AI→AO can be represented by a positive semidefinite Choi operator M on AI⊗AO.
Using one common convention,
M = Σ_{ij} |i⟩⟨j| ⊗ 𝓜(|i⟩⟨j|).
Complete positivity corresponds to
M≥0.
Trace preservation becomes
Tr_{A_O} M = I_{A_I}
under this unnormalised-maximally-entangled-state convention.
Some references transpose input spaces in the probability formula or normalise the maximally entangled state differently. Process-matrix formulas are convention sensitive; the positivity and causal structure are not.
4. Process matrix probability rule
Represent each local CP event by Choi operators MAa|x and MBb|y.
A bipartite process matrix W is a positive semidefinite operator on
A_I⊗A_O⊗B_I⊗B_O
such that probabilities are given, in a standard Choi convention, by
p(a,b|x,y)=Tr[(M^A_{a|x}⊗M^B_{b|y})W].
W must satisfy additional linear normalisation constraints ensuring that valid trace-preserving local channels produce probabilities summing to one.
5. Why positivity alone is not enough
If W≥0, the trace rule gives nonnegative probabilities for positive local Choi operators.
But probabilities must also satisfy
Σ_{a,b}p(a,b|x,y)=1
for every pair of trace-preserving local instruments.
This imposes linear projector/partial-trace constraints on W.
A mathematically positive operator outside that affine subspace is not a valid process matrix.
6. Definite order A≺B
A process compatible with A≺B must forbid signalling from B’s later choice y to A’s earlier outcome statistics.
Operationally,
p(a|x,y)=p(a|x)
for every local instrument choice at B.
B may depend on information originating at A, so A→B signalling is allowed.
The process-matrix subspace encoding this one-way signalling condition is the operator counterpart of an ordinary circuit wire from A to B.
7. Causal separability
A bipartite process is causally separable if it can be written
W_sep=qW^{A≺B}+(1−q)W^{B≺A}, 0≤q≤1.
This includes a classical random switch that chooses one causal order per run.
A process that cannot be written in this form is causally nonseparable.
The definition generalises in multipartite settings, where recursive causal orders and conditioning on earlier outcomes require more care.
8. Classical mixture versus coherent control
Suppose a classical coin sends a target through A then B on heads and B then A on tails.
If the coin is ignored, the process is a probabilistic mixture of definite orders and therefore causally separable.
If instead a quantum control qubit coherently controls the two routes and retains phase coherence between them, interference between orders can become observable.
That coherent-order resource underlies the quantum switch.
9. Quantum switch
The quantum switch takes two operations A and B acting on a target system and a control qubit.
Conditioned on control |0⟩, the target experiences B∘A. Conditioned on |1⟩, it experiences A∘B.
For control state
|+⟩=(|0⟩+|1⟩)/√2,
the two orders are coherently superposed at the higher-order process level rather than classically sampled. Chiribella and collaborators introduced the switch as a higher-order quantum transformation and identified computational/communication advantages for selected tasks. [2]
10. Worked unitary-switch state
Let A and B be unitaries UA,UB acting on target |ψ⟩.
The ideal switch maps
|+⟩_c|ψ⟩_t → [|0⟩ U_BU_A|ψ⟩ + |1⟩ U_AU_B|ψ⟩]/√2.
Measure the control in the ± basis. The unnormalised target branches are proportional to
(U_BU_A+U_AU_B)|ψ⟩for +;(U_BU_A−U_AU_B)|ψ⟩for −.
The control therefore interferes the two orders and can reveal commutator/anticommutator structure.
11. Commuting versus anticommuting example
If UA and UB commute,
U_BU_A−U_AU_B=0.
The − control outcome has zero probability ideally.
If they anticommute,
U_BU_A+U_AU_B=0,
so the + branch vanishes.
This illustrates one task where coherent control of order can extract a global relation using one invocation of each unknown operation under ideal oracle assumptions.
12. Higher-order quantum maps
An ordinary quantum channel maps states to states.
A superchannel maps channels to channels while preserving complete positivity under extension.
Quantum combs describe ordered networks with open slots into which operations can be inserted.
The quantum switch is naturally viewed as a higher-order map whose inputs are operations A and B and whose output is a channel on target plus control.
Process matrices generalise the probability description even further, allowing valid local-operation statistics without presupposing one global acyclic circuit order.
13. Causal games
Oreshkov, Costa and Brukner introduced a bipartite guessing game in which A and B receive random inputs and must communicate information under a task bit deciding which direction matters. [1]
Any strategy compatible with a definite causal order satisfies a success-probability bound—the causal inequality.
The process-matrix framework permits valid processes whose correlations exceed that bound.
This demonstrates logically that local quantum mechanics plus globally definite causal order is a stronger assumption than local quantum mechanics alone within the formalism.
14. Causal inequality
A causal inequality is analogous in spirit to a Bell inequality.
It constrains observed probabilities p(a,b|x,y,…) under the assumption that events admit a definite causal order, possibly chosen probabilistically.
A violation is device-independent with respect to internal quantum descriptions in the same sense that the conclusion rests on observed correlations plus declared causal-game assumptions rather than a trusted tomography of W.
But the assumptions differ from Bell locality; causal inequalities concern order and signalling structure, not spacelike local hidden variables.
15. Causal nonseparability does not imply causal-inequality violation
A causally nonseparable W may nevertheless generate correlations compatible with a definite-order causal model for every available local instrument in a given scenario.
The quantum switch is the key example: it is causally nonseparable, yet in standard bipartite device-independent scenarios its correlations do not simply furnish the original Oreshkov–Costa–Brukner causal-inequality violation.
Therefore:
- causal witness → trusted-device certification of nonseparability;
- causal inequality → device-independent statistical certification of indefinite causal order under its assumptions.
16. Causal witnesses
The set of causally separable process matrices is convex.
Therefore, by convex separation, a causally nonseparable W can be detected by a Hermitian witness S such that
Tr(SW_sep)≥0
for all causally separable Wsep, while
Tr(SW)<0
for the target process.
Witness optimisation can be formulated as a semidefinite program because positivity and process-validity constraints are semidefinite/linear.
17. Semidefinite-program structure
Process-matrix questions often reduce to convex optimisation:
- W≥0;
- linear normalisation constraints L(W)=W;
- causal-separability cones formed from A≺B and B≺A subspaces;
- linear witness objective Tr(SW).
This makes causal structure mathematically close to entanglement and steering theory, where convex sets and separating witnesses also organise resources.
18. Generalised robustness of causal nonseparability
One can quantify how much arbitrary valid-process noise must be mixed into W before it becomes causally separable.
Define robustness R through the smallest r≥0 for which
(W+rW_noise)/(1+r)
is causally separable.
This is analogous to robustness measures of entanglement, coherence and contextuality.
The precise value depends on which noise processes are allowed in the optimisation.
19. Experimental quantum-switch implementations
Photonic experiments have encoded the control in path or another degree of freedom so that one branch applies A then B and the coherent branch applies B then A.
Interference of the control demonstrates coherence between the two circuit-order branches.
Certifying causal nonseparability requires more than seeing two paths. Experiments must relate observed statistics to a witness/model excluding classical mixtures of fixed order under stated assumptions.
20. Communication advantages
Indefinite-order resources can improve selected communication, discrimination and query tasks.
Examples include discriminating whether two unknown unitaries commute or anticommute, and improving information transmission through certain noisy-channel combinations.
These advantages are task specific. They do not imply that an indefinite-order computer universally outperforms every fixed-order quantum circuit.
21. Indefinite order is not time travel
Process matrices relax assumptions about global ordering of local operations. They do not automatically create closed timelike curves, permit messages to one’s own past or violate logical consistency.
Valid processes are constrained precisely so local probabilities remain normalised and nonnegative.
Claims about spacetime geometry require an additional physical theory connecting the operational process to gravitational causal structure.
22. Quantum causal models
Another research direction generalises classical causal Bayesian networks to quantum systems.
Classical models factor a joint distribution according to a directed acyclic graph.
Quantum causal models replace conditional probability tables by channels/processes and must handle interventions, quantum common causes, entanglement and incompatible measurements.
Fixed-DAG quantum causal models and indefinite-order process matrices overlap conceptually but are not identical formalisms.
23. Common cause versus direct cause
Correlations between A and B can arise because:
- A influences B directly;
- B influences A directly;
- a common source influences both;
- mixtures or coherent higher-order structures combine these possibilities.
Quantum causal inference asks which interventions and observations distinguish these mechanisms.
Entanglement can make observational correlation stronger than classical intuition while still requiring careful intervention logic to infer causation.
24. Interventions matter
Correlation alone does not identify causal direction.
An intervention replaces a local mechanism by a chosen operation and observes downstream changes.
Process-matrix laboratories are already operational intervention slots: the experimenter may choose arbitrary local CP maps subject to the scenario.
This is why the formalism is naturally causal rather than merely correlational.
25. Purification of processes
Ordinary mixed quantum states can be purified by embedding them in a larger Hilbert space.
Analogously, one may ask whether a process matrix arises as a reduction of a larger pure process/higher-order unitary structure.
Not every mathematically valid process has an obvious standard-circuit realisation. Physical realisability is therefore a stronger question than formal process validity.
26. Common misconception: a random switch is an indefinite causal order
A classical random choice between A≺B and B≺A is causally separable. Indefinite causal order in the resource sense requires something that cannot be represented as such a probabilistic mixture.
27. Common misconception: the quantum switch necessarily violates a causal inequality
The switch is causally nonseparable, but causal-inequality violation is a stronger device-independent statement. The two concepts should not be conflated.
28. Common misconception: indefinite causal order means indefinite event time
A process concerns causal relations among local operations. Laboratory events can still occur at well-defined spacetime locations in particular implementations while the effective order in which operations act on the target is coherently controlled.
29. Worked synthesis problem
A control qubit coherently selects the order of two unknown unitaries U and V acting on target |ψ⟩.
Step 1: Prepare control. Use |+⟩.
Step 2: Switch output.
[|0⟩VU|ψ⟩+|1⟩UV|ψ⟩]/√2.
Step 3: Measure control in ± basis. The target amplitudes become proportional to (VU±UV)|ψ⟩.
Step 4: If UV=VU. The minus branch vanishes.
Step 5: If UV=−VU. The plus branch vanishes.
The experiment uses coherent interference between alternative orders. A classical 50:50 mixture of UV and VU would erase the relative phase and could not perform this same interference test.
30. Practice set
- What is meant by A≺B?
- What mathematical object represents one local CP map in the process formalism?
- Write the process-matrix probability rule schematically.
- Why must W satisfy more than W≥0?
- Define causal separability in the bipartite case.
- What distinguishes a classical random order from the quantum switch?
- What do ± measurements of the switch control reveal?
- What is a causal inequality?
- Why does causal nonseparability not automatically imply causal-inequality violation?
- What is a causal witness?
- Why are semidefinite programs useful here?
- Why should indefinite causal order not be described automatically as time travel?
Answers
- A’s laboratory can causally influence B while B’s later choice cannot signal to A’s earlier statistics.
- A Choi operator.
p(a,b|x,y)=Tr[(M^A_{a|x}⊗M^B_{b|y})W]in the stated Choi convention.- It must also satisfy linear normalisation constraints so arbitrary local CPTP maps yield total probability one.
W=qW^{A≺B}+(1−q)W^{B≺A}with 0≤q≤1.- A classical mixture has one definite order per run; the switch preserves coherent phase between order branches.
- Interference proportional to the sum or difference of ordered compositions, exposing anticommutator/commutator structure.
- A bound on observed correlations satisfied by every process with definite causal order under the specified causal-game assumptions.
- Operator-level nonseparability can exist even when the available correlations admit a definite-order causal model.
- A Hermitian observable/linear functional nonnegative on all separable processes and negative on a target nonseparable process.
- Process validity, separability and witness optimisation are expressed with positive-semidefinite cones plus linear constraints.
- The formalism changes causal-order assumptions among operations, not automatically spacetime chronology or logical consistency.
Sources and further study
[1] Ognyan Oreshkov, Fabio Costa and Časlav Brukner, Quantum correlations with no causal order, Nature Communications 3, 1092 (2012). The foundational process-matrix formalism and causal-inequality violation.
[2] Giulio Chiribella, Giacomo Mauro D’Ariano, Paolo Perinotti and Benoît Valiron, Quantum computations without definite causal structure, Physical Review A 88, 022318 (2013). The quantum switch and higher-order computational framework.
[3] Mateus Araújo and colleagues, Witnessing causal nonseparability, New Journal of Physics 17, 102001 (2015). Causal witnesses, semidefinite programming and quantum-switch certification.
[4] Cyril Branciard and colleagues, The simplest causal inequalities and their violation. A systematic treatment of device-independent causal inequalities.
Batch 15 series navigation
- Guide 57: Quantum Amplitude Estimation, Approximate Counting, Monte Carlo and Quadratic Sampling Speedups
- Guide 58: Quantum Chemistry Mathematics, Electronic Structure, Fermionic Hamiltonians and Qubit Mappings
- Guide 59: Measurement-Based Quantum Computation, Qubit Graph States, Cluster States and Feed-Forward
- Guide 60: Quantum Causal Models, Process Matrices, Causal Inequalities and Indefinite Causal Order
- Return to the BTT Mathematics Learning Hub
Educational note: “indefinite causal order” has several operational and mathematical meanings across the literature. Claims should specify whether they concern causal nonseparability, a causal witness, a causal inequality, a quantum-switch implementation or a spacetime model.