sorab
commented
8 years ago Let us look at flow between 2 nodes. If head at the upstream node is below the bottom of the sill between the two nodes then the flow from the upstream node should be zero. Thus, no water goes AWAY from the node so head should not drop at that node. So, why would heads consistently drop at each iteration when head was below the bottom (the adjustment was just bringing it closer to bottom)? The culprit was line IF(EKR.LT.1.0E-7) EKR = 1.0E-7 in subroutine KR_CAL which gave that slight conductance when it should not have done so, thus sending water away from the dry node and reducing the head, which we promptly adjust back up to the bottom (almost) and that process repeats itself. Commenting out this line does the trick.
Regarding Hnew(N) = HTEMP(N)(1.0-closebot) + Bot(N)closebot), this is what was intended. If HNEW is below bottom then we are already below bottom. That is why the adjustment is to use HTEMP (which was above bottom) multiplied by a factor of 0.1 with 0.9*Bot to bring it even closer to the bottom (with reduced Kr and thus hopefully a less head reduction due to outflow from the upstream node when it is getting dry. If you use HNEW, it is already below the bottom and that is not what is intended. If head is too much below the bottom, then when there is rewetting of that node, you get a big jump in the head value (up to the bottom and then the additional physical part) and that is why I try to have it as close to the bottom as possible when CONSTANTCV is not used.
What we need in USG is what Rich has put into NWT which is to look at the bottom of the domain (all layers) and not let heads go below that whether CONSTANTCV or not.
Why it does converge when you allow heads to go below the bottom is that the numbers (Kr times the absolute K value) are small enough that the adjustment is less than the tolerance. I know you put the tolerance very small here to check this issue so I am not sure what to say here. I will play with this some more to see if I can have an answer to this.
We can discuss all this face-to-face if it is a bit confusing as there are so many little things involved here.
rniswon
langevin-usgs
I added a new problem to my usg repository. The problem is called test046_flowdivert (https://github.com/OGW-USGS/mfusg/tree/develop/test-cmp/test046_flowdivert). The NWT model is also in there. It is a 2D model and the bottom surface rises in the middle of the domain. Constant heads cause flow from left to right, and the flow is diverted around the hump in the middle of the domain. See figure below.
This problem solves fine with mfnwt, but bottom averaging must be used. The problem solves fine with mf6, with or without bottom averaging. But the problem does not solve with mfusg, with or without damping (and thus bottom averaging). It appears that in the dry cells, the linear solver puts the head below the bottom and then the bottom averaging pulls the head back up. If I shut off the bottom averaging in USG by commenting it out (Hnew(N) = HTEMP(N)_(1.0-closebot) + Bot(N)_closebot), then it solves fine.
Joe and I talked a bit, and its not clear how best to fix this. The problem has a good mass balance, even though it is not technically converged. Perhaps convergence should be evaluated differently in this case. It is also unclear why you are using HTEMP in the bottom averaging part. It seems like that should be hnew, although that doesn't help for this problem.