From Real Space to Field Theory: Understanding RG Limitations

Recommended resources

The freely available resources this lesson will direct you to.

Hello! Welcome back to our study of the Renormalization Group.

In our last lesson, we developed a powerful conceptual picture of universality. We saw how the RG transformation creates flows in a vast "theory space" of Hamiltonians. The critical behavior of a system is dictated not by its microscopic starting point, but by the universal, scale-invariant fixed point to which it flows. This elegantly explains why diverse systems exhibit identical critical exponents.

Today, we shift our focus from the "why" to the "how." We will critically assess the real-space RG methods we've used, like block-spin transformations and decimation. While these methods provide a tangible illustration of RG ideas, they suffer from significant practical and theoretical problems. Our learning outcome for this lesson is to identify the limitations of real-space RG methods that motivate a field-theoretic approach. Understanding these limitations is the crucial bridge to the more powerful and systematic framework of momentum-space RG, which we will begin exploring in the next lesson.

1. The Fundamental Problem: Proliferation of Couplings

The core idea of real-space RG is to integrate out some degrees of freedom and find a new Hamiltonian for the remaining ones. In one dimension, this worked perfectly. The new Hamiltonian for the remaining spins had the same form as the original, just with a renormalized coupling constant.

Does this simplicity survive in higher dimensions? Let's consider the 2D Ising model on a square lattice. Imagine we integrate out the spins on every other site in a checkerboard pattern.

A 2D square lattice with a checkerboard pattern indicating spins to be integrated out (e.g., black) and spins to be kept (e.g., white).

Caption: A checkerboard decimation scheme for the 2D Ising model. Integrating out the 'black' spins will induce new interactions between the remaining 'white' spins.

When we sum over a "black" spin $σ_{B}$ , it connects its four "white" neighbors ( $σ_{W 1}, σ_{W 2}, σ_{W 3}, σ_{W 4}$ ). The partition function contains a term like:

σ_{B} = \pm 1 \sum e^{K σ_{B} (σ_{W 1} + σ_{W 2} + σ_{W 3} + σ_{W 4})} = 2 cosh [K (σ_{W 1} + σ_{W 2} + σ_{W 3} + σ_{W 4})]

If you expand this hyperbolic cosine, you will find it generates not only pairwise interactions between the white spins (like $σ_{W 1} σ_{W 2}$ ), but also four-spin interactions ( $σ_{W 1} σ_{W 2} σ_{W 3} σ_{W 4}$ ) and a host of other complex terms.

This is the central difficulty of real-space RG: even if you start with a simple nearest-neighbor Hamiltonian, the very first RG step generates a new, much more complicated Hamiltonian with next-nearest-neighbor, four-spin, and even longer-range interactions. This is known as the proliferation of interactions. The simple form of the Hamiltonian is not preserved under the RG transformation, and the calculation quickly becomes intractable.

The following reading illustrates this problem and gestures toward the alternative we will eventually adopt.

A Short Introduction to Renormalisation Group Methods

This section of a guide by Steven Thomson vividly illustrates the practical issues with real-space RG in 2D and sets the stage for why a momentum-space approach is often preferred.

Please read the section that starts with 'Now how would we apply this scheme in two dimensions…?' until the introduction of momentum-space RG. Notice how simple decimation schemes either fail to capture the 2D physics or create awkward new lattice geometries. This highlights the ad-hoc nature of these methods.

2. "Solutions" as Uncontrolled Approximations

Since an exact real-space RG transformation is generally impossible in $d > 1$ , we are forced to make approximations. These approximations are often "uncontrolled," meaning there is no small parameter that justifies them. Their accuracy can be difficult to assess without knowing the exact answer beforehand.

Let's examine two famous approximation schemes to see their strengths and, more importantly, their weaknesses.

Case Study 1: The Niemeijer-van Leeuwen Cumulant Expansion

This is a more systematic approach for a 2D Ising model on a triangular lattice. The key idea is to partition the Hamiltonian $H$ into a solvable part $H_{0}$ (interactions within blocks of spins) and a perturbation $U$ (interactions between blocks). One then performs a cumulant expansion in $U$ . This is still an approximation because we truncate the expansion, usually at first or second order, even though the inter-block coupling is not necessarily small near criticality.

The following video walks through this calculation. Your goal is not to absorb every algebraic detail, but to appreciate the logic of the approximation and, crucially, to see the final comparison with exact results.

14. Position Space Renormalization Group, Part 2

Professor Mehran Kardar from MIT provides a detailed walkthrough of the Niemeijer-van Leeuwen approximation. This exemplifies a more sophisticated, yet still limited, real-space RG scheme.

Please watch two segments from this lecture: The Method (4:51 - 34:35): Skim this part to understand the setup. Notice how the Hamiltonian is split and a 'majority rule' is used to define the new spins. The derivation of the recursion relation for the coupling K' is based on a first-order approximation. The Result & Limitations (40:13 - 49:47): This is the key part. Pay close attention as the professor calculates the critical exponent y_t \approx 0.88 and compares it to the exact 2D Ising value of y_t=1. He explicitly calls this an 'uncontrolled approximation' and notes the immense difficulty of going to the next order.

As you saw, while the method gives a non-trivial result that is a decent first guess, it is not exact. Improving it is computationally prohibitive, and its accuracy is not guaranteed.

Case Study 2: Migdal-Kadanoff Bond Moving

This is a much simpler, but also more transparently ad-hoc, approximation. To avoid generating complex couplings when decimating, one simply "moves" problematic bonds onto other bonds, strengthening them in the process. It's a clever trick, but it's not derived from first principles.

Let's see how this "quick and dirty" method fares.

14. Position Space Renormalization Group, Part 2

In the same lecture, Professor Kardar demonstrates the Migdal-Kadanoff bond-moving approximation. This highlights how far one can get with a clever, albeit uncontrolled, simplification.

Watch the segment from 49:47 to 1:02:13. Observe the elegant simplicity of the bond-moving trick. Note the resulting critical exponent y_t \approx 0.75, which is worse than the previous method's result. Crucially, the professor mentions that this approximation gets progressively worse in higher dimensions.

The Migdal-Kadanoff scheme is easy to implement but provides a less accurate result for the 2D Ising model. Its uncontrolled nature is even more apparent.

3. A Summary of Limitations

Having worked through these examples, we can now summarize the main limitations of real-space RG methods.

Proliferation of Couplings: In dimensions $d > 1$ , RG transformations generate an infinite number of new, more complex interactions, making exact solutions impossible.
Uncontrolled Approximations: To manage this proliferation, we resort to approximations like truncation or ad-hoc schemes (bond-moving). These are "uncontrolled" as there is no small parameter to justify them, and their reliability is hard to assess. As a review on the topic notes, "there are equally plausible approximations that yield poor results."
Lattice-Dependence: The specific construction of the RG transformation (the size and shape of blocks, the decimation rule) is highly dependent on the underlying lattice geometry. This is cumbersome and conceptually at odds with universality, which states that critical behavior should be independent of such microscopic details.
Lack of Generality: These methods are naturally designed for discrete lattice models. It is not obvious how to apply them to systems described by continuous fields (like the liquid-gas transition) or to study dynamic critical phenomena.
No Clear Path to Systematic Improvement: For methods like Migdal-Kadanoff, it is not clear how one could systematically improve the calculation. While one could in principle include more terms in a cumulant expansion, the algebraic complexity explodes, making this impractical.

4. The Path Forward: Momentum-Space RG

The limitations of real-space RG are not a dead end. Instead, they point us toward a more powerful, systematic, and universal framework. This framework is built on two key ideas:

Move to a Field Theory: Near a critical point, where the correlation length diverges, the microscopic lattice details become irrelevant. The universal physics is captured by the long-wavelength fluctuations of a continuous order parameter field, $ϕ (x)$ . This is the basis of Landau-Ginzburg theory.
Work in Momentum Space: For a continuous field, it is natural to work in Fourier space. The degrees of freedom are the Fourier modes $\tilde{ϕ} (k)$ . In this language, "coarse-graining" takes on a precise and systematic meaning: integrating out the high-momentum ("fast") modes that correspond to short-wavelength fluctuations.

This momentum-space RG approach, pioneered by Kenneth Wilson, overcomes the limitations of real-space RG:

It is formulated for a continuous field theory, making it independent of any specific lattice. Universality is built in from the start.
It allows for controlled perturbative calculations, most famously the $ϵ$ -expansion, where $ϵ = 4 - d$ serves as a small parameter.

Test your understanding!

Imagine you are tasked with calculating the critical exponents for a 3D Heisenberg model (where spins are 3D vectors, not just up/down).

Why would a real-space RG approach be even more challenging for this model than for the Ising model?
Now, consider a momentum-space RG approach based on a field theory. Why might this approach be more straightforward, conceptually, for handling the continuous symmetry of the Heisenberg spins?

Show answer

A real-space RG approach would be significantly harder for the Heisenberg model. The "proliferation of couplings" problem would be even more severe. When integrating out a vector spin, the new interactions generated between the remaining vector spins would involve not just dot products, but more complex tensor couplings. Keeping track of the flow of all these new coupling constants would be a nightmare. Furthermore, defining a simple "majority rule" for vector spins is not as straightforward as for Ising spins.
A momentum-space RG approach starts by identifying the appropriate order parameter field. For the Heisenberg model, this is a three-component vector field $ϕ (x)$ . The Landau-Ginzburg-Wilson Hamiltonian would be constructed based on symmetry principles—in this case, rotational $O (3)$ symmetry. The RG procedure would then involve integrating out high-momentum modes of this field. The entire process works with continuous fields and their symmetries from the beginning, avoiding the clumsy, lattice-specific blocking and decimation steps. This makes it a much more natural and generalizable framework for systems with continuous symmetries.

Conclusion

In this lesson, we have seen that while real-space RG is a powerful conceptual tool that gives us a tangible feel for coarse-graining, its practical application is fraught with difficulties. The proliferation of interactions in dimensions greater than one forces us into uncontrolled, lattice-dependent approximations whose reliability is uncertain.

Key Takeaways:

The primary limitation of real-space RG is the proliferation of interactions, where each RG step generates a more complicated Hamiltonian.
To cope with this, we use uncontrolled approximations (like truncation or ad-hoc schemes) that are not systematically improvable and can give inaccurate results.
Real-space methods are also lattice-dependent and not easily generalized to continuous systems.
These limitations motivate a move to momentum-space RG, a field-theoretic approach that works with continuous order parameter fields in Fourier space. This framework is more universal and allows for controlled, systematic calculations.

Preview of the Next Lesson:

Our discussion has naturally led us to the doorstep of field theory. To begin our journey into momentum-space RG, we first need to learn how to translate our discrete lattice models into the language of continuous fields. In the next lesson, we will do just that, focusing on the crucial first step: outlining the mapping of the Ising model to a $ϕ^{4}$ field theory via the Hubbard-Stratonovich transformation. This will provide us with the starting point for the powerful Wilsonian RG calculations to come.