Trust radius reduction for "Unexpected energy increase" of negligible size
Closed this issue · 0 comments
ptrab commented
During the optimization, it makes sense to change the trust radius to make only steps of reasonable size. However, I often find this happens even for minor energy changes (see below). Do you think it is necessary even at this level?
I have no deep understanding of it, but I assume this could make the calculation take longer to converge, as the trust radius is kept very small for, in principle, converged geometries. By this, the steps needed until convergence should increase unnecessarily. Maybe there can also be a threshold for ΔE that prevents trust radius updates when they are below.
Spent 0.0 s preparing the first cycle.
cycle Δ(energy) max(|force|) rms(force) max(|step|) rms(step) s/cycle
0 nan* 0.010059 0.001391 0.023344 0.006090 127.27
Unexpected energy increase (0.013892 au)! Trust radius: old=0.3, new=0.1
1 0.013892* 0.010074 0.001385 0.011261 0.003219 242.93
2 -0.014758* 0.006472 0.000929 0.016833 0.004543 248.26
Unexpected energy increase (0.014178 au)! Trust radius: old=0.1, new=0.1
3 0.014178* 0.007396 0.001071 0.020433 0.005226 243.66
4 -0.000184* 0.006858 0.001000 0.022375 0.005105 128.25
5 -0.014071* 0.007472 0.001088 0.038674 0.010192 252.57
Unexpected energy increase (0.013652 au)! Trust radius: old=0.2, new=0.1
6 0.013652* 0.005838 0.000870 0.021090 0.005083 250.15
7 -0.000129* 0.005540 0.000829 0.048443 0.010231 128.39
8 -0.013254* 0.008967 0.001335 0.064303 0.017690 246.95
9 -0.000588* 0.007893 0.001197 0.065005 0.018942 133.31
-----------------------------------------------------------------------------------
10 -0.001065* 0.003824 0.000624 0.024803 0.006966 137.67
11 -0.000335* 0.001316 0.000240 0.017524 0.004525 133.47
12 -0.000076* 0.000775 0.000120 0.022380 0.005079 132.30
13 -0.000051* 0.000414 0.000077 0.008418 0.001803 130.90
14 -0.000019* 0.000247 0.000050 0.004693 0.001176 130.47
15 -0.000012* 0.000140 0.000026 0.003756 0.000846 129.01
16 -0.000001* 0.000117 0.000018 0.003164 0.000731 128.12
Unexpected energy increase (0.000004 au)! Trust radius: old=0.4, new=0.1
17 0.000004* 0.000072 0.000012 0.002473 0.000607 126.17
Unexpected energy increase (0.000004 au)! Trust radius: old=0.1, new=0.1
18 0.000004* 0.000041 0.000008* 0.002775 0.000705 127.32
Unexpected energy increase (0.000003 au)! Trust radius: old=0.1, new=0.1
19 0.000003* 0.000052 0.000009* 0.002546 0.000685 128.06
Unexpected energy increase (0.000001 au)! Trust radius: old=0.1, new=0.1
-----------------------------------------------------------------------------------
20 0.000001* 0.000057 0.000009* 0.001880 0.000546 127.35
21 -0.000000* 0.000042 0.000007* 0.001039 0.000297 128.03
22 -0.000001* 0.000017 0.000004* 0.000565 0.000124 126.56
Unexpected energy increase (0.000000 au)! Trust radius: old=0.1, new=0.1
23 0.000000* 0.000008* 0.000002* 0.000216 0.000049 125.93
Unexpected energy increase (0.000000 au)! Trust radius: old=0.1, new=0.1
24 0.000000* 0.000008* 0.000001* 0.000115 0.000027* 124.94
Unexpected energy increase (0.000001 au)! Trust radius: old=0.1, new=0.1
25 0.000001* 0.000006* 0.000001* 0.000085 0.000021* 123.65
Unexpected energy increase (0.000000 au)! Trust radius: old=0.1, new=0.1
26 0.000000* 0.000003* 0.000000* 0.000064 0.000016* 124.76
Converged!