Abstract: We describe a method to extract force and diffusion parameters from single trajectories of Brownian particles based on the principle of maximum likelihood. The analysis is well-suited for out-of-equilibrium trajectories, even when a limited amount of data is available and the dynamical parameters vary spatially. We substantiate this method with experimental and simulated data, and discuss its practical implementation, strengths, and limitations.
Introduction
Analyzing the motion of Brownian particles—tiny objects buffeted by random molecular collisions—is a cornerstone of microscale physics, allowing researchers to measure forces and fluid properties. Traditionally, these measurements rely on statistical methods like Boltzmann statistics, which require the system to be in equilibrium, or "big data" approaches that necessitate thousands of trajectories to build reliable histograms.
These requirements pose a significant hurdle for studying out-of-equilibrium systems or small, fragile particles like proteins. This paper introduces a Maximum Likelihood Analysis (MLA) framework designed to extract precise physical parameters from single, relatively short, and noisy trajectories.
Methodology: Maximum Likelihood Analysis
Instead of local data binning, the proposed MLA method performs a global fit of the entire trajectory. The approach is rooted in the Smoluchowski equation, which describes particle motion under the influence of a force ($F$) and diffusion ($D$). For short time intervals ($\Delta t$), the displacement of a particle follows a Gaussian distribution. The mean of this distribution is determined by the drift velocity (force), while the variance is determined by the diffusion coefficient.
The "Likelihood" represents the probability that a specific set of physical parameters would produce the observed sequence of steps. By maximizing the log-likelihood, researchers can identify the parameters that best describe the observed trajectory. This method is particularly robust for handling spatial variations; by using functional forms to describe how $F$ and $D$ vary with position, the global fit can optimize parameters across the entire path.
Experimental Validation and Results
The method was validated using both simulated data and experimental trajectories of paramagnetic spheres. In one experiment, a 2.8 $\mu$m paramagnetic bead was driven by a constant magnetic field gradient in water.
Key results from these validations include:
- Extreme Data Efficiency: MLA proved to be roughly 200 times more data-efficient than traditional variance sampling. In one instance, MLA required only 1,400 timepoints to achieve the same accuracy that traditional methods needed 300,000 timepoints for.
- High Precision: For the single paramagnetic bead, MLA estimated a force of $59 \pm 2$ fN and a diffusion coefficient of $6.28 \pm 0.15 \times 10^{-14}$ m²/s, matching conventional results with significantly less data.
- Mapping Complex Interactions: The method successfully mapped the $1/r^4$ force law of magnetic dipoles and distance-dependent diffusion coefficients caused by hydrodynamic coupling near surfaces.
Discussion and Limitations
While powerful, MLA is dependent on the choice of an appropriate functional form for the force and diffusion profiles. An incorrect physical model can lead to systematic errors. Additionally, "trajectory blurring" caused by finite camera exposure times can lead to overestimations of force if the exposure time is not significantly smaller than the frame time step.
Despite these considerations, MLA is remarkably resilient. Even when high-force regions of a trajectory are removed, the global fit remains accurate within a few percent. This makes it an ideal tool for single-molecule tracking, where trajectories are often cut short by photobleaching.
Conclusion
Maximum Likelihood Analysis offers a robust, data-efficient alternative to traditional Brownian motion analysis. By leveraging the entire trajectory through a global fit, it unlocks the ability to study molecular-level interactions and out-of-equilibrium dynamics that were previously obscured by noise or data scarcity.
Summary updated