A video processing method includes determining, for a conversion between a block of a video and a bitstream representation of the video, that an output value from an inverse secondary transform with a reduced dimension (e.g., an inverse low frequency non-separable transform) is constrained within a range of [min, max] inclusively. The inverse secondary transform is applicable to the block between a de-quantization step and an inverse primary transform. The reduced dimension is reduced from a dimension of the block, and min and max are integer values. The method also includes performing the conversion based on the determining.