Inferring causal structure: a quantum advantage

Katja Ried, Megan Agnew, Lydia Vermeyden, Dominik Janzing, Robert W. Spekkens, Kevin J. RescharXiv 2014 · DOI · OpenReview

TL;DR

This paper introduces "causal tomography," a new quantum method for inferring causal relationships between systems. Remarkably, it demonstrates that quantum correlations, specifically entanglement and coherence, can reveal causal structures even without direct intervention, a feat impossible in classical systems.

What Problem Does This Paper Address?

The fundamental challenge of inferring causal relationships from observed correlations is a cornerstone of scientific inquiry. Classically, observing a correlation between two variables, say A and B, doesn't tell us if A causes B, or if a third factor, C, causes both A and B. To resolve this ambiguity, we typically need to perform randomized interventions – actively manipulating one variable and observing the effect on the other. This paper asks: can quantum mechanics offer a different approach? Specifically, can the unique properties of quantum correlations help us disentangle causal structures, perhaps even without the need for active interventions?

Methods

The authors introduce a novel technique called causal tomography. This method extends conventional quantum tomography to infer not just the state of a quantum system, but also its causal relationship with another system. They propose two schemes:

  1. Interventionist Scheme: This is the quantum analogue of a randomized trial. By performing interventions (randomized preparations) on an initial system (let's call it D), and then measuring the subsequent systems (B and C), they can gather enough data to fully reconstruct the "causal map" – a mathematical description of how D influences B and C. This allows for a complete solution to the causal inference problem.
  2. Passive Observation Scheme: This is where the quantum advantage truly shines. Here, no active interventions are performed on the initial system. Instead, the system is allowed to evolve naturally after a measurement. The authors show that, surprisingly, quantum correlations alone can sometimes reveal the causal structure. This is possible when the underlying causal relationships are limited to specific forms, such as a probabilistic mixture of a unitary process (direct cause) and a pure, maximally entangled state (common cause).

The experimental implementation involves using polarization-entangled photons and a probabilistic swap gate to control the causal relationship between two optical modes.

Datasets

Not discussed in the paper. The paper focuses on a theoretical framework and experimental demonstration using a controlled quantum system.

Key Insights and Ideas

  • Quantum Advantage in Causal Inference: The most significant insight is that quantum mechanics provides an advantage in causal inference. Unlike classical systems where passive observation is insufficient, quantum correlations can reveal causal structures.
  • Causal Tomography: This new framework unifies and generalizes existing tomography techniques, allowing for the complete characterization of causal relationships.
  • Entanglement and Coherence are Key: The paper highlights that entanglement and coherence, hallmarks of quantum information processing, are not just interesting phenomena but are necessary for this quantum advantage in causal inference. Without them, distinguishing causal structures via passive observation becomes impossible.
  • Passive Observation Sufficiency (Under Promise): While full causal inference requires intervention, the paper demonstrates that under specific "promises" about the nature of the causal structure (e.g., a mixture of a unitary process and an entangled state), passive observation alone is sufficient to solve the problem.

Key Results

  • The authors successfully implemented a quantum optical experiment to control and probe causal relationships between two optical modes.
  • They demonstrated that causal tomography, using the interventionist scheme, can reconstruct the causal map with high fidelity (average fidelity of 94.5%).
  • Crucially, they showed that the passive observation scheme could also achieve high fidelity reconstructions (average fidelity of 97.6%) for specific causal structures, outperforming classical limitations.
  • The experiment confirmed that entanglement and coherence are essential for this quantum advantage.

Limitations and Future Work

  • Promise of Extremality: The ability to solve the causal inference problem with passive observation alone relies on a "promise" about the nature of the causal structure (e.g., it's a mixture of a unitary process and a pure entangled state). Without this promise, passive observation alone is insufficient.
  • General Causal Mechanisms: The paper focuses on specific causal structures. A key future direction is to explore what can be inferred about more general causal mechanisms from passive observations alone.
  • Scaling to More Systems: The current work focuses on pairs of systems. Generalizing these schemes to arbitrary numbers of systems is an open and important research question.
  • Non-Markovian Dynamics: The authors suggest that their methods could be used to test whether the dynamics of open quantum systems are Markovian or non-Markovian, as non-Markovian evolution can be viewed as a common-cause mechanism between the system and its environment.

Summary updated

Abstract

The problem of using observed correlations to infer causal relations is relevant to a wide variety of scientific disciplines. Yet given correlations between just two classical variables, it is impossible to determine whether they arose from a causal influence of one on the other or a common cause influencing both, unless one can implement a randomized intervention. We here consider the problem of causal inference for quantum variables. We introduce causal tomography, which unifies and generalizes conventional quantum tomography schemes to provide a complete solution to the causal inference problem using a quantum analogue of a randomized trial. We furthermore show that, in contrast to the classical case, observed quantum correlations alone can sometimes provide a solution. We implement a quantum-optical experiment that allows us to control the causal relation between two optical modes, and two measurement schemes -- one with and one without randomization -- that extract this relation from the observed correlations. Our results show that entanglement and coherence, known to be central to quantum information processing, also provide a quantum advantage for causal inference.
BibTeX
@misc{ried2014inferring,
  author = {Katja Ried and Megan Agnew and Lydia Vermeyden and Dominik Janzing and Robert W. Spekkens and Kevin J. Resch},
  title = {{Inferring causal structure: a quantum advantage}},
  year = {2014},
  eprint = {1406.5036},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1406.5036},
}
Figure