1: \begin{abstract}

2: We demonstrate a novel technique to obtain singular-value decomposition (SVD) of 

3: the coupled-cluster triple excitations amplitudes, $t_{ijk}^{abc}$. The presented method is based on 

4: the Golub-Kahan bidiagonalization strategy and does not require $t_{ijk}^{abc}$ to be stored. 

5: The computational cost of the method is comparable to several CCSD iterations. Moreover, the number 

6: of 

7: singular vectors to be found can be predetermined by the user and only those singular vectors which 

8: correspond to the largest singular values are obtained at convergence. We show how the subspace of 

9: the most important singular vectors obtained from an approximate triple amplitudes tensor can be 

10: used to solve equations of the CC3 method. The new method is tested for a set of small and 

11: medium-sized molecular systems in basis sets ranging in quality from double- to quintuple-zeta. It 

12: is 

13: found that to reach the chemical accuracy ($\approx 1$ kJ/mol) in the total CC3 energies as little 

14: as $5-15\%$ of SVD vectors are required. This corresponds to the compression of the $t_{ijk}^{abc}$ 

15: amplitudes by a 

16: factor of ca. $0.0001-0.005$. Significant savings are obtained also in calculation of interaction 

17: energies or rotational 

18: barriers, as well as in bond-breaking processes.

19: \end{abstract}

20: