1: \begin{abstract}

2: 

3: Using a set of 28 high resolution, high signal to noise ratio (S/N) 

4: QSO Ly$\alpha$ absorption spectra, we investigate the non-Gaussian features 

5: of the transmitted flux fluctuations, and their effect upon the 

6: power spectrum of this field. We find that the spatial distribution of the 

7: local power of the transmitted flux on scales $k \geq 0.05$~s/km is highly 

8: spiky or intermittent. The 

9: probability distribution functions (PDFs) of the local power are long-tailed. 

10: The power on small scales is dominated by small probability events,

11: and consequently, the uncertainty in the power spectrum of the transmitted 

12: flux field is generally large. This uncertainty arises due to the 

13: slow convergence of an intermittent field to a Gaussian limit required 

14: by the central limit theorem (CLT). To reduce this uncertainty, it 

15: is common to estimate the error of the power spectrum by selecting 

16: subsamples with an ``optimal" size. We show that this conventional 

17: method actually does not calculate the variance of the original 

18: intermittent field but of a Gaussian field. Based on the analysis of 

19: intermittency, we 

20: propose an algorithm to calculate the error. It is based on a bootstrap 

21: re-sampling 

22: among all independent local power modes. This estimation doesn't require 

23: any extra parameter like the size of the subsamples, and is sensitive to the 

24: intermittency of the fields. This method effectively reduces the uncertainty 

25: in the power spectrum when the number of independent modes matches the 

26: condition of the CLT convergence.

27: 

28: \end{abstract}

29: