In lattice field theory, calculating the fundamental properties of particles requires extracting clear signals from quantum correlation functions. Ideally, these correlation functions act as a filter that isolates the lowest-energy state of a system—its ground state. However, this filtering process is inherently messy. The raw data generated by lattice simulations is frequently polluted by higher-energy states, a phenomenon known as excited-state contamination.
Uncontrolled excited-state contamination introduces systematic biases into calculations of critical physical constants, such as the nucleon axial coupling ($g_A$), or the phase shifts in particle scattering. To achieve percent-level precision, physicists cannot simply wait for these unwanted states to disappear; they must actively design mathematical frameworks to suppress them.
The Catch-22 of Euclidean Time
To understand why excited-state contamination is so challenging, we must look at how correlation functions behave in Euclidean time ($t$). A typical lattice correlation function $C(t)$ can be expressed as a sum of decaying exponentials:
$$C(t) = \sum_{n=0}^{\infty} A_n e^{-E_n t} = A_0 e^{-E_0 t} + A_1 e^{-E_1 t} + A_2 e^{-E_2 t} + \dots$$
Here, $E_0$ represents the ground-state energy, while $E_1, E_2, \dots$ represent the energies of progressively higher excited states. Because higher energies cause their corresponding exponentials to decay faster, the ground state ($e^{-E_0 t}$) will eventually dominate the signal if we look at a large enough value of $t$.
This setup implies a straightforward solution: simply measure the correlation function at very late times. Unfortunately, researchers run directly into a severe statistical barrier known as the signal-to-noise problem. As N. Ishii et al. observe in Phase shifts in $I=2$ ${\pi}{\pi}$-scattering from two lattice approaches, multi-particle and multi-nucleon measurements suffer from a rapidly degrading signal-to-noise ratio at later times.
This creates a difficult trade-off:
- At early times, the statistical noise is low, but the signal is heavily contaminated by higher-energy excited states that have not yet decayed (Evan Berkowitz et al. in Calm Multi-Baryon Operators).
- At late times, the excited states have vanished, but the statistical noise completely swallows the ground-state signal.
If this contamination is not properly managed, it skews physical predictions. For instance, Chia Cheng Chang et al. demonstrate in both Nucleon axial coupling from Lattice QCD and A per-cent-level determination of the nucleon axial coupling from quantum chromodynamics that failing to isolate the ground state introduces significant systematic errors when computing fundamental parameters like $g_A$.
Signal Quality
^
| [Early Time] [Late Time]
| High Statistical Precision Clean Ground State
| But Heavy Excited-State Pollution But Overwhelming Statistical Noise
| (Spectral Pollution) (Signal-to-Noise Problem)
| \ /
| \___________________________________/
+----------------------------------------------------> Euclidean Time (t)
Disentangling States via Multi-Operator Basis Expansion
If we cannot afford to wait for excited states to die out naturally, we must find a way to mathematically disentangle them at earlier times. The most effective way to do this is by expanding the basis of interpolating operators used to create and annihilate states on the lattice.
If a simulation relies on a single, simplistic operator, it lacks the resolution to distinguish between the ground state and nearby excited states. This limitation causes the simulation to conflate the states, leading to severe discrepancies between lattice results and theoretical expectations, as detailed in the 2019 study Studies of I=0 and 2 $\pi\pi$ scattering with physical pion mass.
By expanding the operator basis to include a diverse mix of single-particle and multi-particle structures, researchers can construct a matrix of correlation functions. Solving a Generalized Eigenvalue Problem (GEVP) across this matrix allows them to isolate individual states.
A prominent example of this strategy is found in pion-pion ($\pi\pi$) scattering and kaon decay simulations. Christopher Kelly, in a series of papers including Update on the improved lattice calculation of direct CP-violation in K decays and S-wave pi-pi I=0 and I=2 scattering at physical pion mass, showed that incorporating a specific scalar "sigma" ($\sigma$) operator into the basis successfully projects out contaminating excited states. This addition resolved long-standing discrepancies in moving-frame calculations and aligned lattice phase shifts with theoretical predictions.
Once a robust operator basis is established, researchers apply simultaneous multi-state spectral fitting. As Christopher Kelly notes in Update on the improved lattice calculation of direct CP-violation in K decays and Yu Meng shows in A new method for a lattice calculation of $\eta_c\rightarrow 2\gamma$, fitting multiple distinct operators simultaneously acts as a mathematical filter, separating the pure ground-state signal from higher-energy artifacts even when the individual channels are noisy.
(A conceptually parallel approach exists in quantum chemistry: Sebastian Keller et al. demonstrate in An efficient matrix product operator representation of the quantum chemical Hamiltonian that targeting excited states through direct orthogonalization within a Matrix Product State framework avoids the optimization blurring common in state-averaging methods.)
Hierarchical Optimization and "Calm" Operators
The excited-state problem becomes even more acute in multi-baryon systems, such as a pair of interacting nucleons. When multiple nucleons are initialized in close spatial proximity, they experience significant inelastic distortion, generating massive excited-state contamination (Evan Berkowitz et al. in Progress in Multibaryon Spectroscopy).
To solve this, a hierarchical optimization strategy was developed based on a simple insight: a multi-particle system cannot have a clean ground-state signal if its constituent building blocks are noisy. In Calm multi-baryon operators, the authors identify that single-nucleon excited-state contamination is actually the dominant source of noise in two-nucleon correlation functions at late times.
Traditional Approach:
Raw Quarks ---> Standard Nucleon Operators ---> Multi-Nucleon Correlator (High Noise)
"Calm" Operator Approach:
Raw Quarks ---> Matrix Prony Optimization ---> "Calm" Nucleons ---> Clean Multi-Nucleon Correlator
The "calm baryon" method solves this problem through a multi-step process:
- Isolate the Constituents: Before constructing the multi-baryon system, the individual single-nucleon operators are optimized using statistical techniques like the Matrix Prony method.
- Strip Single-Particle Noise: This optimization filters out the excited-state contamination of the individual nucleons, isolating their pure ground states.
- Combine and Displace: These pre-cleaned, "calm" nucleon operators are then combined and spatially displaced at the source.
As Evan Berkowitz et al. demonstrate in Calm multi-baryon operators and Calm Multi-Baryon Operators, this combination yields exceptionally clean multi-nucleon correlation functions. Because the constituent operators are already optimized, the ground-state signal can be reliably extracted at much earlier Euclidean times. By reading the signal early, physicists bypass the late-time exponential decay of the signal-to-noise ratio entirely, enabling high-precision calculations of nuclear properties that were previously computationally intractable.
Key Foundational Work
Phase shifts in $I=2$ ${\pi}{\pi}$-scattering from two lattice approaches (2014) — contributed 1 concept elements to this direction.
An efficient matrix product operator representation of the quantum chemical Hamiltonian (2015) — contributed 1 concept elements to this direction.