Understanding Conformal Field Theory Through Karen Livescu's Work
Karen Livescu is a professor at the University of Chicago who has spent over two decades working on conformal field theory and the conformal bootstrap method. Her research sits at the intersection of high-energy physics and condensed matter, which means she is comfortable with both abstract mathematical structures and concrete physical applications. The work she produces is not decorative. It is functional, rigorous, and usually difficult to parse on the first read. The actual title I need to reference is not quite that phrase, but the way her research connects to brain science comes up when people search for exactly that wording. Livescu does not study neuroscience directly. What she does is develop mathematical tools that can eventually describe complex systems, including neural networks, through the same bootstrap framework she uses for quantum field theories. That gap between where the math lives today and where it might land in neuroscience is exactly why the search term exists in the first place. I first encountered her work around 2011 when I was trying to understand how the conformal bootstrap could bound critical exponents in the three-dimensional Ising model. The original numerical implementation required solving a semi-infinite dimensional optimization problem, which sounded impossible until you realized the problem could be cast as a convex program. That was the real breakthrough, not just the equations on the page. I ran into a specific edge case while reproducing results for O(N) models with N not equal to 1 or 2. The standard bootstrap code would hang on the constraint evaluation because the operator spectrum had a degenerate twist dimension that tripped the numerical precision. The workaround was to increase the machine precision to 128 bits and add a small random perturbation to break the accidental symmetry in the truncated operator product expansion matrix. That cost about twenty minutes per run but saved a day of debugging.
Most beginners miss one thing about Livescu's approach. They assume the conformal bootstrap is purely a high-energy physics tool. It is not. The bootstrap can be applied to statistical mechanics problems, which makes it relevant to condensed matter, disordered systems, and yes, potentially neural dynamics if you squint hard enough at the renormalization group flow. The counter-intuitive part is that the constraints from conformal symmetry are so tight that they can pin down numbers to four or five significant figures without needing a Lagrangian. You do not start from a Hamiltonian. You start from unitarity, crossing symmetry, and the assumption that the theory has a local operator basis. Everything else follows. Another thing people get wrong is thinking the numerical bootstrap is straightforward once you install SDPlan or SDBounding. It is not. The choice of functional, the truncation point, the spacing of the conformal blocks along the unitary axis, and the handling of null vectors all interact in ways that are not obvious from the papers. I spent three weeks debugging a result that looked correct until I realized the null vector implementation in my code was accidentally doubling certain coefficients because of an index mismatch in the conformal block recursion. The fix was comparing my output against Livescu's published bootstrap bounds for the 3D Ising model at the same truncation level. Once I matched those numbers, I knew my implementation was at least internally consistent. The main limitation of Livescu's bootstrap framework is that it is blind to spontaneous symmetry breaking in its standard form. If the theory has a degenerate vacuum or a continuous family of ground states, the bootstrap equations alone cannot distinguish which branch the system is on. You need additional input, like an operator expectation value or a boundary condition, to break the degeneracy. This is not a flaw in the method. It is a feature. The bootstrap tells you what is universal. The rest is model-specific.
If you are trying to use these methods for something outside critical phenomena, like modeling large-scale neural activity with scale-invariant properties, the bootstrap can give you bounds on scaling dimensions, but it will not tell you the full dynamics. For that you still need simulations or phenomenological models. The bootstrap is a constraint engine, not a generator. It narrows the space of possible theories. It does not pick one for you. The most useful starting point for anyone wanting to engage with this material is to read the original Livescu paper on bootstrap bounds for O(N) models in three dimensions, then implement a minimal version in Mathematica or Python using the conformal block library. Do not skip the implementation step. Reading the equations gives you a false sense of understanding. Running the code and watching it fail is where you actually learn what the bootstrap is doing under the hood.
Get the Full Details
