The Evolution of Occam’s Inversion: From 1987 to Modern Geophysics

In 1987, geophysicists Steven C. Constable, Robert L. Parker, and Catherine G. Constable published a seminal paper titled "Occam’s Inversion: A practical algorithm for generating smooth models for the inversion of electromagnetic data." While the primary focus was on electromagnetic sounding, the mathematical framework provided a strong solution to the universal problem of non-uniqueness in geophysical data. By applying the philosophical principle of Ockham’s Razor—that the simplest explanation is usually the best—the authors introduced a regularization method that prioritized "smoothness" over high-frequency oscillations in model parameters.

This methodology has since become a cornerstone in the interpretation of seismic surface wave data. In contemporary geophysics, researchers and engineers use Occam’s Inversion to derive 1D shear wave velocity (Vs) profiles from dispersion curves. The transition from rigid, few-layer models to highly discretized, smooth profiles allows for a more realistic representation of geological gradations. This approach prevents the introduction of spurious features or artifacts that often result from attempts to over-fit noisy experimental data.

What changed

The publication of the 1987 algorithm marked a significant departure from the standard inversion practices of the mid-20th century. Before this shift, geophysical inversion typically relied on discrete, over-simplified models or unstable unconstrained solutions. The following changes occurred as the "smoothness" philosophy gained traction:

• Shift from Discrete Layers to Continuous Variation:Instead of assuming a subsurface composed of three or four thick, homogeneous layers, researchers began using models with dozens of thin layers where properties changed gradually.
• Introduction of Regularization:The algorithm introduced a Lagrange multiplier to balance the fit between the observed data and the model's roughness. This allowed for a controlled trade-off between statistical accuracy and physical plausibility.
• Reduction of Interpretive Bias:By automating the selection of the simplest model that fits the data, the algorithm reduced the need for the practitioner to "guess" the number of layers, a process that frequently led to inconsistent results between different analysts.
• Computational Stability:Occam’s Inversion provided a stable numerical framework for iterative linearized least-squares problems, which had previously been prone to divergence when dealing with ill-posed equations.

Background

The fundamental challenge of geophysical inversion lies in the fact that many different subsurface models can produce the same set of surface observations. This "non-uniqueness" is particularly prevalent in surface wave analysis, where Rayleigh and Love wave dispersion data represent an integrated effect of the subsurface properties. Traditionally, if a geophysicist wanted to find a model that fit the data within a 5% error margin, they might find hundreds of complex, jagged models that met the criteria. However, many of these models would contain extreme velocity variations that were geologically improbable.

The concept of regularization was not entirely new in 1987—Tikhonov and Miller had explored similar mathematical territories—but the Constable et al. Paper made the concept "practical" for the computing power available at the time. The authors argued that since we cannot know the true complexity of the earth from limited data, the most honest representation is the one that is as smooth as possible. In this context, "smooth" usually means minimizing the first or second derivative of the velocity with respect to depth.

Mathematical Formulation of Smoothness

The Occam approach involves minimizing a specific functional. Rather than simply minimizing the sum of the squared residuals (the difference between observed and calculated data), the algorithm minimizes a combination of the residual and the "roughness" of the model. The objective function is typically expressed as:

"Find the modelMThat minimizes R(m) = ||&partial²m||² subject to ||d - f(m)||² ≤ X²_Target"

In this equation,MIs the vector of model parameters (such as shear wave velocity),DIs the observed data,F(m)Is the forward modeling operator, andX²_TargetIs the desired level of misfit. By iteratively adjusting the Lagrange multiplier, the algorithm finds the maximum smoothness possible while staying within the error bounds of the data.

Application to Surface Wave Profiles

In the discipline of Surface Wave Hub research, this inversion technique is applied to the dispersion curves of Rayleigh waves. Rayleigh waves are dispersive, meaning different frequencies penetrate to different depths and travel at different speeds. By measuring the phase velocity at various frequencies, a dispersion curve is generated. Inverting this curve reveals the shear wave velocity profile of the ground.

Transitioning from Rigid Models

Early methods for Spectral Analysis of Surface Waves (SASW) often used a limited number of layers (e.g., 5 to 10). The analyst would manually adjust the thickness and velocity of each layer until the theoretical curve matched the field data. This "layer-stripping" or manual fitting was prone to over-fitting, where the analyst would add a thin, high-velocity layer just to match a single noisy data point on the dispersion curve. Occam’s Inversion removed this subjectivity by using a large number of layers of fixed thickness and letting the smoothness constraint dictate the velocity gradient.

Impact on Lithological Characterization

The use of smooth profiles has improved the ability of geophysicists to characterize transition zones, such as the weathering of bedrock or the gradual compaction of sedimentary soils. While a rigid layered model might show a sharp boundary at five meters, an Occam-derived profile might show a gradual increase in velocity from four to six meters. This gradient is often a more accurate representation of the physical state of the subsurface, where moisture content and pressure change continuously rather than abruptly.

Modern Software Implementation

Today, the legacy of the 1987 paper is found in numerous geophysical software packages. Tools such asGeopsy(widely used for Microtremor Array Measurements or MAM) andWinSASW(developed for controlled-source testing) incorporate regularization parameters that descend directly from the Occam philosophy.

Geopsy and the Neighborhood Algorithm

While Geopsy’sDinverModule often utilizes a "Neighborhood Algorithm" (a stochastic, global search method), it frequently allows users to apply constraints that mirror the smoothness requirements of Occam’s Inversion. The software enables the definition of parameter ranges that prevent unrealistic velocity inversions (where a deeper layer is significantly slower than a shallower one) unless the data strongly demands it. This prevents "ringing" or oscillations in the resulting shear wave velocity profiles.

WinSASW and Regularization

In engineering applications, WinSASW provides a structured environment for inverting SASW data. It allows for the adjustment of regularization weights, enabling the engineer to decide how much to trust the data versus how much to enforce a smooth profile. In cases of high-quality data with low noise, the regularization can be relaxed; in cases of high environmental noise, the smoothness constraint is tightened to ensure the resulting model remains physically interpretable for infrastructure assessment.

Challenges and Limitations

Despite its ubiquity, Occam’s Inversion is not a universal solution for every geological setting. Because the algorithm explicitly penalizes roughness, it can "smear" sharp geological contacts. If a site contains a distinct concrete slab over soft clay, or a sharp soil-bedrock interface, a strictly smooth model will show a transition zone instead of a crisp boundary. In such cases, the very feature the engineer is looking for might be dampened by the smoothness constraint.

To address this, modern implementations often allow for "breaks" in smoothness. If a user has prior knowledge (e.g., from a borehole log) that a sharp interface exists at a certain depth, they can decouple the smoothness constraint at that specific point, allowing the algorithm to find a smooth model on either side of a known discontinuity.

Conclusion

The evolution of Occam’s Inversion from a 1987 theoretical paper to a daily tool in engineering and seismology reflects the geophysical community's move toward more objective, repeatable data analysis. In the study of seismic surface waves, the ability to generate reliable 1D profiles without the interference of human bias or numerical instability has fundamentally improved subsurface imaging. Whether investigating the foundations of a bridge or detecting buried utilities, the application of "practical" smoothness remains a vital bridge between mathematical theory and geological reality.