When Imaginary Time Becomes Real: A Partial Construction of Timelike Liouville
29 May 2026, Yanjiang
A cylinder geometry bridges Euclidean and Lorentzian quantum gravity through analytic continuation of timelike Liouville field theory.
What would it take to build a quantum theory of gravity on a cylinder? Not the gently curved space of our universe, but two-dimensional quantum gravity where time itself runs along a circle — a miniature toy cosmos where every ripple of geometry is a quantum fluctuation. The question is not idle. Timelike Liouville field theory has been a candidate for such a theory for decades, a mathematical chalice that promises to capture positive curvature quantum gravity in a tractable form. Yet every attempt to make the theory rigorous in real, Lorentzian time has stumbled. The necessary mathematical objects — correlation functions, local observables, the very notion of a vacuum — have remained out of reach, locked behind a wall of analytic continuation so forbidding that most researchers stayed on the Euclidean side, where time is imaginary and life is easier.
Sourav Chatterjee, working alone from the Department of Statistics at Stanford University, has now produced what amounts to a map of the terrain — incomplete in places, daring in others, but undeniably the first systematic attempt to cross from imaginary time to real time for this notoriously sensitive theory. His preprint (arXiv:2605.29203) constructs Lorentzian correlation functions for a restricted class of observables on the cylinder, proves that spacelike separated operators commute (at least in a limited sense), and assembles an algebraic quantum field theory — one that notably lacks a positive-definite inner product and therefore does not yield a Hilbert space of physical states. The construction is a partial success, an elegant feat of analytic machinery, and a vivid reminder of how much remains unknown.
The core challenge is deceptively simple to state. In Euclidean quantum field theory, one works with imaginary time, where the path integral is a well-behaved probabilistic object — essentially a theory of random surfaces. For timelike Liouville, the Euclidean correlation functions have been known in the Coulomb gas formalism for years. They involve integrals of vertex operators over the worldsheet, with screening charges inserted to ensure charge neutrality. To go to real time, one must analytically continue these Euclidean correlators in the time variables, mapping the cylinder’s imaginary-time interval onto the real line. That sounds like a routine operation, but for timelike Liouville it is anything but: the integrands have branch cuts, the screening contour must be deformed in intricate ways, and the resulting Lorentzian correlators must be shown to exist as genuine mathematical objects.
Chatterjee’s key insight is to restrict attention to the “integer screening sector,” where the total screening charge is an integer. In this sector, the analytic continuation can be carried out explicitly using contour integral formulas. The Lorentzian correlators emerge as boundary values of analytic functions, defined by smearing vertex operators against test functions along a cycle in the complex time plane. The result is a set of numbers that behave, formally, like vacuum expectation values of a local quantum field theory: they depend only on the ordering of the time variables and are invariant under cylinder translations. A locality theorem shows that vertex operators at spacelike separation commute — a cornerstone of any consistent quantum field theory. So far, so promising.
But here the dialectical tension begins to mount. The integer screening sector is, by any measure, a vanishingly small subset of the full theory. It is a needle’s eye through which only a tiny fraction of the theory’s physical content passes. The full timelike Liouville theory should contain operators with arbitrary, non‑integer screening charges; restricting to integer values is like studying quantum electrodynamics with only neutral particles — interesting, but hardly the whole story. Earlier work on compactified imaginary Liouville (Guillarmou et al., arXiv:2310.18226) sharpens a pivotal question: is the integer sector rich enough to reconstruct the full OPE algebra, or is it a separate, exactly solvable subsector disconnected from the deeper quantum geometry?
The second tension is even more unsettling. Locality, as proven in this preprint, applies to vertex operators. But the smeared observables that generate the full algebra — built from integer‑charge fields — are not embedded in the standard Tomita–Takesaki modular framework that underpins axiomatic quantum field theory, because the vacuum form is indefinite and no GNS Hilbert space exists. The algebraic framework Chatterjee assembles uses an indefinite inner product (for a coupling constant smaller than roughly 0.354, the Hermitian form is explicitly indefinite), meaning there is no Hilbert space, no state‑operator correspondence in the usual sense, and no guarantee that the theory’s predictions are probabilistic. This is not a failure of consistency — the construction is mathematically precise — but it challenges the assumption that a quantum theory of gravity must be built on the bedrock of positive probabilities. The probabilistic construction of non‑compactified imaginary Liouville by Usciati et al. (arXiv:2505.09390) sharpens this question: is there a genuine probability measure behind the Lorentzian correlators, or is the indefinite inner product a fundamental feature that cannot be eliminated?
Perhaps the most profound ambiguity lurks in the analytic continuation itself. The contour definition for Lorentzian correlators is not unique: different choices of integration cycle in the complex time domain can yield different boundary values. Chatterjee selects a specific cycle that yields the expected vacuum expectation values in the integer sector, but the physical meaning of this choice remains obscure. Is it the only one that respects locality, or does it reflect a deeper, hidden constraint that the theory must satisfy? The bootstrap community, represented here by Mühlmann et al. (arXiv:2505.08890), would naturally ask whether this ambiguity can be resolved by demanding consistency with the full set of three‑point functions of the timelike Liouville CFT. Until such constraints are known, the Lorentzian construction sits on a scaffold whose base is the author’s choice of contour, not a self‑evident principle.
Yet the paper achieves something remarkable. It shows that a substantial fragment of the Euclidean‑to‑Lorentzian reconstruction program survives even when the inner product is indefinite and the representation is not on a Hilbert space. The algebraic net of local algebras, the continuous cyclic representation, and the translation automorphism all exist and satisfy isotony and locality, albeit in a weakened algebraic sense. For theorists who have long wondered whether timelike Liouville can be made sense of as a quantum field theory at all, this partial construction is a demonstration that the standard machinery of axiomatic quantum field theory can be adapted — not straightforwardly, but with a kind of stubborn elegance — to a setting that has resisted it for years.
What this work challenges, more than any specific technical claim, is the unspoken assumption that a consistent quantum gravity theory must be a unitary quantum field theory on a Hilbert space. Timelike Liouville, if it exists, may require a broader notion of physical state: a vector space with an indefinite metric, where probabilities emerge only after gauge‑fixing or restricting to a positive‑definite subspace. This is not unprecedented — the Gupta–Bleuler method in QED does something similar — but in quantum gravity the stakes are higher. If the fundamental degrees of freedom of spacetime cannot be accommodated in a Hilbert space, then the architecture of quantum theory itself may need rethinking.
Chatterjee has given us a path, not a cathedral. The unanswered questions — the restriction to the integer sector, the contour ambiguity, the missing probability measure — are not flaws that invalidate the construction. They are precise scientific questions that now, for the first time, can be asked within a mathematically controlled framework. Whether the integer sector can be embedded into a full theory, whether the contour can be fixed by physical principles, and whether the indefinite inner product can be tamed are not mysteries to be lamented; they are the next hills to climb. In the quiet conversation between imaginary time and real time on that cylinder, we have just heard the first clear sentence — and we are only beginning to learn its grammar.
— Yanjiang
Yanjiang is an online editor of LoomSci.com.
References
- Sourav Chatterjee, A Lorentzian construction of timelike Liouville field theory on the cylinder, arXiv:2605.29203
- Guillarmou et al., Compactified Imaginary Liouville Theory, arXiv:2310.18226
- Mühlmann et al., On the three-point functions in timelike N=1 Liouville CFT, arXiv:2505.08890
- Usciati et al., Probabilistic construction of non compactified imaginary Liouville field theory, arXiv:2505.09390