Non-polarizing beam splitter is an optical component widely used for beam splitting. It is primarily designed to divide incident light into two beams with specified reflection and transmission ratios while maintaining the original polarization state of the input light. Thanks to its low polarization discrimination， minimal stress birefringence， and strong polarization-maintaining capability， non-polarizing beam splitters are extensively employed in optical systems such as space remote sensing， projectors， and Michelson interferometers. Although an ideal non-polarizing beam splitter is designed to preserve the polarization state of the incident light， practical non-polarizing beam splitters exhibit certain non-ideal polarization characteristics due to inherent limitations in materials and coating designs. Moreover， these polarization properties are highly sensitive to both the angle and direction of the incident light. In some complex optical systems， a non-polarizing beam splitter may need to simultaneously process light incident from both forward and reverse directions， inevitably operating under reverse incidence. Therefore， a systematic study of the polarization properties of the non-polarizing beam splitter under reverse incidence is of great significance for high-precision optical measurement and polarization-sensitive systems. Currently， most research focuses on the influence of factors such as temperature and angular deviation on the polarization performance of the non-polarizing beam splitter， while studies on polarization characteristics under reverse incidence remain limited. This paper addresses scenarios in complex optical systems where the non-polarizing beam splitter must be used in reverse. We systematically analyze the polarization properties of the non-polarizing beam splitter under different incident directions from three perspectives： fundamental principles， simulation analysis， and experimental validation. The variations in polarization characteristics of both reflected and transmitted channels with respect to incident wavelength are analyzed and compared， and the mechanism underlying the influence of incident direction on the polarization properties of the non-polarizing beam splitter is revealed.

To achieve this objective， a combined theoretical， numerical， and experimental approach was adopted. Numerical simulations were performed using Essential Macleod thin-film design software to construct a realistic model of a non-polarizing beam splitter coating stack. The model enabled quantitative calculation of wavelength-dependent polarization parameters， including diattenuation and phase retardation， for both reflected and transmitted channels under forward and reverse incidence conditions. The simulations covered the visible spectral range and assumed identical angles of incidence for both illumination directions to isolate the effect of direction reversal. To validate the simulation results， high-precision experimental measurements were conducted using Mueller matrix ellipsometer. Commercial non-polarizing beam splitter samples （models CCM1-BS013/M and CCM5-BS016/M） were selected as representative specimens. The complete 4×4 Mueller matrices of the samples were measured over a wavelength range from 380 nm to 800 nm under both forward and reverse illumination. Subsequently， the Lu-Chipman decomposition method was applied to the measured Mueller matrices to extract key polarization parameters， specifically diattenuation and phase retardation， for the reflected and transmitted channels. Multiple measurements were performed to ensure repeatability and to assess sample-to-sample consistency.

The simulation results reveal a pronounced dependence of polarization characteristics on the direction of incident light， particularly for the reflected channel of the non-polarizing beam splitter. Under forward and reverse incidence， substantial discrepancies were observed in both diattenuation and phase retardation of the reflected beam. Specifically， the maximum change in phase retardation between the two incidence conditions reached approximately 53.64°， with an average variation on the order of 29.49°， while the maximum difference in diattenuation was approximately 0.321， with an average difference of about 0.027 3. These findings indicate that reversing the illumination direction significantly alters the polarization behavior of the reflected channel. In contrast， the transmitted channel exhibited negligible variation in both diattenuation and phase retardation across the same conditions. Experimental measurements corroborated these trends with strong qualitative and quantitative agreement. Mueller matrix ellipsometry data showed that， for the reflected channel， the maximum difference in diattenuation between forward and reverse incidence was approximately 0.11， with an average difference of around 0.03， while the maximum difference in phase retardation reached approximately 51.5°， with an average difference of about 28.76°. Importantly， repeated experiments on different non-polarizing beam splitter samples consistently demonstrated the same directional dependence： significant polarization variation in the reflected channel and nearly identical polarization behavior in the transmitted channel under both illumination directions.