+
We study non-parametric calibration of local volatility models, which is formulated as an inverse problem of partial differential equations with Tikhonov regularization. In contrast to the existing literature minimizing the distance between theoretical and market prices of options as a calibration criterion, we instead minimize the distance between theoretical and market implied volatilities, complying with market practices. We prove that our calibration criterion naturally leads to the well-posedness of the calibration problem. In particular, comparing to Jiang and Tao (2001), we obtain a global uniqueness result, where no additional weight functions are required. Numerical results reveal that our method achieves a better trade-off between minimizing calibration errors and reducing overfitting.