The Flumen of Spectral Form: Emergent Geometry in Non-Fermi Liquids

The strange metal phase in high-temperature superconductors and heavy-fermion compounds has long resisted explanation within Landau's Fermi liquid framework. The spectral function—the single-particle excitation spectrum measured by photoemission—shows a broad, incoherent continuum with no sharp quasiparticle peak, and its temperature and frequency scaling follows power laws that defy the standard ω² + T² form. Over the past decade, a radical idea has gained traction: that this anomalous spectral shape is a signature of emergent geometry—a hidden curved spacetime in the low-energy theory. The flumen of spectral form, the flow of spectral weight across energy and momentum, may encode the curvature of an internal anti-de Sitter (AdS) space or the fractal geometry of a Sachdev-Ye-Kitaev (SYK) model. This article is for theorists who already understand the basic puzzle and need a practical comparison of the three main toolkits—holography, SYK, and extended dynamical mean-field theory (EDMFT)—to decide which path to invest in for their next project.

1. The Decision: Who Must Choose and by When

If you are a postdoc or graduate student in theoretical condensed matter physics, the choice of framework for studying non-Fermi liquids is not an abstract preference—it determines which conferences you can publish in, which codes you need to learn, and how long your project will take. The decision typically arises when you have a specific material or model (the two-dimensional Hubbard model near optimal doping, a multi-orbital Anderson lattice, or a SYK-like quantum dot) and need to compute a spectral function that can be compared to angle-resolved photoemission (ARPES) or scanning tunneling microscopy (STM) data. The timeline is driven by funding cycles and thesis deadlines: you usually need a publishable result within 12 to 18 months. The wrong choice can mean spending a year learning a technique that cannot capture the key physics of your system.

Three broad communities have developed distinct approaches. The holographic duality (AdS/CMT) community uses classical gravity in a curved extra dimension to compute retarded Green's functions analytically. The SYK community works with exactly solvable models of randomly interacting fermions that exhibit maximal chaos and a finite entropy at zero temperature. The EDMFT community extends the standard dynamical mean-field theory to include non-local correlations via cluster methods or diagrammatic expansions. Each approach has produced spectral functions that resemble the strange metal data, but they differ in what they assume, what they predict, and how hard they are to implement. This guide helps you map your problem's constraints—microscopic realism, analytical control, numerical cost—onto the right toolkit.

We assume you are comfortable with Matsubara Green's functions, the concept of self-energy, and the basics of strongly correlated electron systems. We will not re-derive Landau's criterion for a Fermi liquid; instead, we focus on the practical trade-offs that determine which method will yield a spectral function you can trust and defend in a review.

2. The Option Landscape: Three Approaches to Emergent Geometry

2.1 Holographic Duality (AdS/CMT)

In this approach, the non-Fermi liquid is described by a quantum field theory that is dual to a classical gravitational theory in one higher dimension. The spectral function is obtained by solving a wave equation in an asymptotically AdS spacetime with a charged black hole. The geometry encodes the scaling properties: the near-horizon region of an extremal black hole gives an emergent IR conformal symmetry, and the spectral function shows power-law behavior with a frequency-dependent exponent. The key advantage is analytical control: for many backgrounds, the retarded Green's function can be written in closed form using hypergeometric functions. The disadvantage is that the microscopic connection to a lattice Hamiltonian is weak—you are effectively assuming that the low-energy theory is a conformal field theory with a large central charge, which is not obviously true for the Hubbard model. Practitioners often start with the simplest bottom-up model (Einstein-Maxwell-dilaton gravity) and then add higher-derivative terms to fit specific scaling exponents.

2.2 Sachdev-Ye-Kitaev (SYK) Models

The SYK model considers N Majorana fermions with random all-to-all interactions of strength J. In the large-N limit, the model is exactly solvable via a saddle-point equation for the Green's function and self-energy. The solution reveals a conformal phase at intermediate energies, with a spectral function that decays as ω^−1/2 at low frequency and a linear-in-temperature resistivity. The model is maximally chaotic, saturating the Maldacena-Shenker-Stanford bound. The geometry here is emergent in the sense that the large-N equations map onto a dilaton-gravity theory in two dimensions, giving a concrete example of holography without supersymmetry. The strength of SYK is that it is a well-defined microscopic model with a controlled expansion in 1/N. The weakness is that the all-to-all interactions are unrealistic for electrons in a crystal—the model lacks momentum dependence, so it cannot describe the momentum-space structure of ARPES data. Recent work extends SYK to lattices (SYK chains or clusters) to reintroduce momentum, but these extensions sacrifice exact solvability.

2.3 Extended Dynamical Mean-Field Theory (EDMFT)

EDMFT starts from a realistic lattice Hamiltonian and maps it to a quantum impurity model self-consistently embedded in a bath. The 'extended' part adds a retarded interaction to capture non-local correlations beyond the single-site approximation. For non-Fermi liquids, one often uses the EDMFT with a cluster extension (DCA or CDMFT) or a diagrammatic extension (GW+EDMFT). The spectral function is computed numerically via continuous-time quantum Monte Carlo (CTQMC) or exact diagonalization of the impurity model. The advantage is microscopic realism: you can input the actual band structure and interaction parameters of a material like cuprates or pnictides. The disadvantage is numerical cost: reaching the low temperatures needed to see the non-Fermi liquid scaling requires significant supercomputer time, and the analytic continuation from imaginary time to real frequencies (via maximum entropy or Padé methods) introduces systematic uncertainty. The emergent geometry here is not a literal spacetime but an effective low-energy theory that may have a holographic dual—the EDMFT self-energy often shows scaling consistent with a local quantum critical point.

3. Comparison Criteria: How to Evaluate the Options

When choosing among these three approaches, you need to weigh at least five criteria: (i) microscopic faithfulness, (ii) analytical control, (iii) computational cost, (iv) ability to produce momentum-resolved spectra, and (v) connection to experimental observables beyond the spectral function. Let us examine each.

Microscopic faithfulness measures how directly the model corresponds to a real material Hamiltonian. EDMFT scores highest here because you can start from a density functional theory (DFT) band structure. Holography scores lowest because it assumes a conformal field theory that may not be the correct UV completion. SYK sits in between: the random coupling is artificial, but the model is a legitimate quantum mechanical system with a tunable interaction strength.

Analytical control is highest for holography (solutions in closed form) and SYK (large-N saddle point), while EDMFT relies on numerical solutions with finite-size and discretization errors. However, analytical control comes at the price of assumptions: holographic models often have free parameters (the dilaton potential, the charge density) that are adjusted to fit data, reducing predictive power.

Computational cost is prohibitive for EDMFT if you need low temperatures (below 0.01 of the bandwidth) and fine frequency resolution. Holography requires solving ODEs, which is cheap. SYK requires solving integral equations, which is moderately cheap for large N but becomes expensive for lattice extensions.

Momentum resolution is essential for comparing to ARPES. Holography can produce momentum-dependent spectral functions by solving the wave equation with a spatial momentum k. SYK in its original form is momentum-independent; lattice SYK models can restore momentum but at the cost of complexity. EDMFT with a cluster can resolve momentum on the scale of the cluster size, which is limited to about 8–16 sites in practice.

Experimental connection beyond the spectral function includes transport (resistivity, Hall coefficient), thermodynamics (entropy, specific heat), and response functions (optical conductivity, Raman scattering). Holography can compute all of these from the same gravitational background. SYK gives the resistivity and entropy directly. EDMFT can compute transport via the Kubo formula but requires additional analytic continuation. The most important criterion for your project may be which observable your experimental collaborators care about most.

4. Trade-Offs: A Structured Comparison

To make the choice concrete, consider a composite scenario: you want to explain the linear-in-temperature resistivity and the universal scaling of the optical conductivity in a strange metal like Bi-2212. Below we compare the three approaches across the five criteria, using a scale of 1 (poor) to 5 (excellent).

The table reveals that no method dominates. Holography offers the best analytical control and low cost, but its weak microscopic connection means that any quantitative fit to data may be post-hoc. SYK provides a controlled microscopic model but lacks momentum resolution unless extended, which then reduces its analytical advantage. EDMFT is the most realistic but is numerically expensive and struggles with low-temperature scaling. The trade-off is essentially between explanatory depth (holography) and microscopic realism (EDMFT).

We recommend that you start by identifying the most important observable for your problem. If you need momentum-resolved spectra at finite temperature and can tolerate a phenomenological model, holography is the fastest path. If you want a controlled calculation of transport and chaos, SYK is ideal. If you must match a specific material's band structure and are willing to invest in large-scale numerics, choose EDMFT. In practice, many groups use a hybrid approach: they fit the scaling exponents from holography or SYK to guide the parameters of an EDMFT calculation.

5. Implementation Path After the Choice

5.1 If You Choose Holography

Start by selecting a bottom-up gravity model. The simplest is the Einstein-Maxwell-dilaton system with a linear dilaton potential, which gives a spectral function with a power-law exponent that depends on the charge density. You will need to solve the wave equation for a probe scalar field in the black hole background. Use a shooting method from the horizon to the boundary, with the boundary condition that the field is ingoing at the horizon. The retarded Green's function is the ratio of the boundary fall-offs. Open-source codes like HoloPy or the AdS/CFT toolkit for Mathematica can handle this. Expect to spend about 2–3 months to generate a spectral function that can be compared to data, including time to learn the numerical ODE solving.

5.2 If You Choose SYK

Begin with the single-site SYK model at large N. Write the self-consistent equation for the Green's function G(τ) and self-energy Σ(τ) in imaginary time. Solve by iteration on a grid of 10⁴–10⁵ points. The spectral function is obtained by analytic continuation using the numerical renormalization group or by solving the real-time Schwinger-Dyson equations on the Keldysh contour. For a lattice extension, start with a one-dimensional chain of SYK dots with nearest-neighbor hopping. The computational cost increases linearly with the number of sites. Plan for 4–6 months to get a momentum-resolved spectral function for a small cluster (up to 8 sites).

5.3 If You Choose EDMFT

Implement the EDMFT self-consistency loop: start with a guess for the hybridization function, solve the impurity model using CTQMC (the TRIQS or iQIST packages are good starting points), compute the local Green's function, extract the new bath, and iterate to convergence. For a cluster, use the DCA approximation with a 4×4 cluster. The main bottleneck is the Monte Carlo sign problem at low temperatures and large interactions. Use the maximum entropy method to analytically continue the self-energy to real frequencies. This is the most time-intensive path: expect 6–12 months to get reliable spectral functions, and be prepared to use a computing cluster with hundreds of cores.

6. Risks If You Choose Wrong or Skip Steps

The most common mistake is to pick a method based on familiarity rather than suitability. A group that exclusively uses EDMFT may spend months trying to reproduce the linear resistivity of a strange metal, only to find that their cluster size is too small to capture the non-local vertex corrections that drive the linear-T behavior. Conversely, a holography enthusiast may fit the spectral function with a beautiful analytic formula, but the model's prediction for the optical conductivity may disagree with experiment because the gravity dual does not include the lattice structure that breaks particle-hole symmetry.

Another risk is underestimating the analytic continuation problem. In EDMFT, the imaginary-time data from CTQMC must be continued to real frequencies, and this step is ill-conditioned. Small noise in the Monte Carlo data can produce spurious peaks in the spectral function. Many published EDMFT spectra have features that are artifacts of the continuation method. To mitigate this, always compare results from two different continuation methods (e.g., maximum entropy and stochastic analytic continuation) and check that the sum rule (integral of the spectral function equals one) is satisfied to within 5%.

A third risk is ignoring the role of disorder. Real materials have impurities that can produce a non-Fermi liquid spectral function through mechanisms unrelated to strong correlations. If your model is translationally invariant, you might misattribute a disorder-broadened peak to emergent geometry. Always check whether your predicted spectral function is robust to adding a small amount of disorder, either by explicit averaging in EDMFT or by including a dissipative term in the holographic model.

Finally, there is the risk of overinterpreting the geometry. The 'emergent spacetime' in holography is a dual description, not a real extra dimension. It is tempting to claim that the strange metal phase is 'dual to a black hole' and leave it at that, but such statements are not falsifiable unless they lead to testable predictions for, say, the butterfly velocity or the energy diffusion constant. If your project does not aim to measure chaos or transport, the geometric language may be a distraction.

7. Mini-FAQ: Common Questions from Practitioners

7.1 Can I use holography to compute the spectral function of the Hubbard model?

Not directly. Holography assumes that the low-energy theory is a conformal field theory, while the Hubbard model has a lattice cutoff and a non-conformal UV. However, you can construct a holographic model that reproduces the scaling of the Hubbard model's self-energy in the intermediate frequency regime. This is a phenomenological mapping, not a derivation. If you need a first-principles calculation, use EDMFT.

7.2 How large does N need to be for SYK to be reliable?

For the spectral function, N = 32 is often sufficient to see the conformal scaling, but the subleading 1/N corrections can be significant for transport. For quantitative comparisons to experiments, aim for N ≥ 64 and compute the 1/N corrections explicitly. The large-N limit is a controlled expansion, but the model's all-to-all coupling means that the physics is effectively infinite-dimensional, so it may miss the effects of spatial dimensionality.

7.3 Why does my EDMFT spectral function show a pseudogap instead of a non-Fermi liquid?

This often happens when the cluster size is too small or the temperature is too high. A pseudogap can be a precursor to the Mott transition, not a genuine non-Fermi liquid. Check the temperature dependence: a true non-Fermi liquid shows a spectral function that fills in as temperature increases, while a pseudogap deepens. Also verify that the self-energy on the real axis has a non-analytic frequency dependence (e.g., Im Σ ∼ ω^1/2), not a simple linear form.

7.4 Can I combine methods? For example, use SYK to generate a self-energy and feed it into a holographic model?

Yes, this is an active area. The SYK model is dual to a two-dimensional dilaton-gravity theory, so it provides a concrete example of holography. You can start from the SYK solution and then add a spatial dimension to create a holographic dual that has momentum resolution. This is technically challenging but has been done for the SYK chain. The hybrid approach is promising for connecting microscopic models to geometric descriptions.

8. Recommendation Recap Without Hype

For a researcher starting a new project on non-Fermi liquid spectral functions, we recommend the following decision process. First, determine whether your primary goal is to explain a specific material's data or to understand a general mechanism. For material-specific work, invest in EDMFT with a cluster of at least 8 sites, and be prepared for a long numerical campaign. For mechanism-focused work, choose SYK if you care about chaos and the low-temperature entropy, or holography if you want to generate a family of spectral functions with tunable exponents quickly.

Second, regardless of your choice, always compute at least two independent observables. A spectral function alone is underdetermined; if your model also reproduces the resistivity and the optical conductivity, the evidence for emergent geometry is much stronger. Third, publish your code and the raw imaginary-time data (for EDMFT) or the gravitational background (for holography) so that others can reproduce your results. The field is moving toward open science, and the most impactful papers are those that provide a complete toolkit.

Finally, keep in mind that the flumen of spectral form—the way spectral weight flows with energy and momentum—is a rich observable that may encode the geometry of the underlying theory. Whether that geometry is a literal extra dimension or an effective description of entangled degrees of freedom, the practical path is to compare multiple theoretical frameworks against the same experimental data. The choice is not permanent: you can start with one method and later validate with another. The key is to make an informed decision now, based on the trade-offs we have outlined, and to execute with discipline. The strange metal problem will not be solved by a single technique, but by the cumulative weight of consistent evidence from many angles.