Scaling Finite Numerical Simulations to the Thermodynamic Limit

16 concepts15 papers1 proposal

In the physical world, macroscopic materials contain an astronomical number of particles—typically on the order of Avogadro's number ($10^{23}$). In theoretical physics, this is treated as the thermodynamic limit, where the system size $L$ and particle number $N$ tend toward infinity while keeping density constant. It is only within this infinite limit that true phase transitions occur as sharp, mathematically non-analytic breaks in a system's properties: a liquid instantly freezes, or a metal abruptly loses its electrical resistance.

However, numerical simulations—whether classical molecular dynamics, quantum Monte Carlo, tensor networks, or neural network quantum states—are fundamentally confined to small, finite geometries. This restriction introduces finite-size effects, creating a persistent bottleneck across condensed matter physics, materials science, and quantum field theory. Understanding why finite size distorts physical reality, and how researchers mathematically bridge the gap between small simulations and the infinite limit, is essential for capturing true bulk physics.

How Finite Scale Distorts Physical Reality

To build intuition for why finite scale poses a problem, imagine a system approaching a continuous phase transition. As the system nears the critical point, fluctuations begin to coordinate over increasingly long distances. The characteristic scale of this coordination is the correlation length, $\xi$. In an infinite system, $\pi$ diverges ($\xi \to \infty$) at the exact transition point, driving the macroscopic divergences seen in physical observables like specific heat or magnetic susceptibility.

In a finite simulation box of linear size $L$, however, the correlation length cannot grow larger than the box itself ($\xi \le L$). The boundaries artificially truncate the long-range correlations. This truncation distorts numerical results in several distinct ways:

Smeared and Shifted Critical Points

Because divergences are capped by the finite boundary, the sharp mathematical singularities of a phase transition are smoothed out into rounded, finite peaks, and the apparent transition point shifts away from its true macroscopic value. As noted by Mohamad Ali Marashli et al. in Unveiling quantum phase transitions by fidelity mapping, finite-size effects cause identified transition points to deviate systematically from theoretical thermodynamic limits. This issue is especially severe in infinite-order transitions, where small system sizes fail to capture the full complexity of long-range correlations. Similarly, Kyungmin Lee et al. demonstrate in Pomeranchuk Instability of Composite Fermi Liquids that critical values for phase transitions derived from finite-particle benchmarks inevitably shift when scaled to macroscopic materials.

Artificial Correlations and Trivialized Landscapes

When a simulation cell is too small, particles or spins interact with their own periodic images or experience artificial constraints that do not exist in bulk matter. In atomistic simulations of lithium manganese oxide spinels, Marco Eckhoff et al. found that phase transition temperatures remain highly sensitive to the dimensions of the simulation cell, even when using large supercells, because small volumes introduce artificial correlations in atomic motion Closing the gap between theory and experiment for lithium manganese oxide spinels using a high-dimensional neural network potential.

In disordered systems like spin glasses, these constraints can completely alter the apparent energy landscape. The authors of Inherent structures in m-component spin glasses demonstrated that simulations on small lattices can prematurely indicate a trivial energy landscape with a single minimum, whereas the true, exponentially complex landscape only emerges when the system size is large enough to decouple finite-size constraints.

The Ambiguity of Phase Stability

Often, a simulation is simply too small to tell whether an observed phase is a stable physical state or a numerical artifact. For decades, researchers have debated whether a 3D spin glass can maintain its ordered state under an external magnetic field. J. Houdayer et al. encountered this bottleneck in Zero-Temperature Responses of a 3D Spin Glass in a Magnetic Field, noting that the properties of small, discrete lattices make it difficult to determine if observed phenomena are genuine phase transitions or scale artifacts. Olivier Martin similarly emphasized in Spin and Link Overlaps in Three-Dimensional Spin Glasses that low-temperature simulations of complex spin systems are frequently restricted to small scales, clouding our understanding of the true thermodynamic limit.

This ambiguity spans multiple domains:

The Mechanics of Finite-Size Scaling

To extract infinite-system physics from finite data, researchers turn to Finite-Size Scaling (FSS) theory. Rather than treating finite-size effects merely as errors, FSS exploits them. By systematically varying the system size $L$ and observing how physical quantities change, researchers can deduce the critical behavior of the infinite system.

Consider a physical quantity $\chi$ (such as magnetic susceptibility) that diverges at a true bulk critical temperature $T_c$ according to a power law in an infinite system:

$$\chi \propto |T - T_c|^{-\gamma}$$

In a finite system of linear dimension $L$, the divergence is cut off because the correlation length $\xi$ cannot exceed $L$. Near the transition, the ratio $\xi/L$ governs the system's behavior. According to scaling theory, the finite-system susceptibility can be modeled by a universal scaling function $f$:

$$\chi(L, T) = L^{\gamma/\nu} f\left( L^{1/\nu} (T - T_c) \right)$$

where $\nu$ is the critical exponent governing the correlation length ($\xi \propto |T - T_c|^{-\nu}$).

Infinite System (L = ∞)             Finite System (L)
       Value                               Value
         ▲                                   ▲
         │     /                             │       .--.
         │    /                              │      /    \
         │   │                               │     /      \
         └───┴────────►                      └────┴────────►
             Tc   Temperature                     Tc   Temperature
     (Sharp Divergence)                      (Smeared, Shifted Peak)

This mathematical relationship provides a powerful recipe for finding the true thermodynamic limit:

  1. Run simulations across a sequence of increasing system sizes (e.g., $L = 4, 8, 12, 16$).
  2. Plot $\chi(L, T) \cdot L^{-\gamma/\nu}$ against $L^{1/\nu} (T - T_c)$.
  3. Adjust the unknown parameters ($T_c$, $\gamma$, and $\nu$) until the distinct curves for every system size collapse perfectly onto a single master curve defined by $f$.

Once this scaling relation is established, evaluating the behavior as $1/L \to 0$ yields the true macroscopic properties. However, this process is demanding. As B. Normand et al. demonstrate in their study of Mott insulator melting, even when explicit efforts are made to extrapolate results to the thermodynamic limit, residual finite-size effects can still degrade accuracy if the underlying simulations fail to reach the asymptotic scaling regime Finite-temperature charge dynamics and the melting of the Mott insulator.

Algorithmic Strategies to Push the Scale

Because finite-size scaling relies on having data from systems large enough to reflect asymptotic scaling laws, a major focus of modern computational physics is developing algorithms that can handle larger system sizes.

Matrix Reduction

For classical lattice models like the 3D Ising model, tracking thermodynamic properties near a transition requires calculating the largest eigenvalues of massive transfer matrices. To make larger scales computationally tractable, Wei Yang et al. developed a matrix reduction technique Simulation of 3-dimensional finite Ising model. By decomposing the standard transfer matrix into a rank-reduced version that preserves the dominant eigenvalues, they made it possible to simulate finite-width 3D systems at scales that capture the divergence of specific heat and magnetization.

Tensor Networks

In quantum field theories, traditional Monte Carlo methods often fail due to the "sign problem"—a phenomenon where quantum interference causes probability weights to become negative or complex. To study the CP-violating Dashen phase in the Schwinger model, Karl Jansen et al. utilized matrix product states (a class of tensor networks) to bypass the sign problem entirely Exploring the <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline"><mml:mrow><mml:mi>C</mml:mi><mml:mi>P</mml:mi></mml:mrow></mml:math>-violating Dashen phase in the Schwinger model with tensor networks. Even with these advanced representations, the authors note that conducting simulations on finite lattices means that achieving a true thermodynamic limit remains an ongoing challenge.

Neural Quantum States

Recently, variational machine learning architectures have emerged as a scalable way to represent complex, many-body wavefunctions. By using neural networks to parameterize the quantum wavefunction, researchers can capture intricate entanglement patterns with fewer parameters than traditional brute-force methods.

Key Foundational Work

Spin and Link Overlaps in Three-Dimensional Spin Glasses (2000) — contributed 1 concept elements to this direction.
Zero-Temperature Responses of a 3D Spin Glass in a Magnetic Field (2001) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_thermodynamic_limit_extrapolation_32595a9d,
  title = {thermodynamic limit extrapolation},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/thermodynamic-limit-extrapolation/rdr_32595a9d4a57482d937a4fbd174d266d},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-22},
}

This exact version (version 1, frozen text):

@misc{lacuna_thermodynamic_limit_extrapolation_32595a9d_v1,
  title = {thermodynamic limit extrapolation},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/thermodynamic-limit-extrapolation/rdr_32595a9d4a57482d937a4fbd174d266d/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-22},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake