How to determine maximum lag.length in VAR?

Question

I am trying to estimate the Vector Autoregression model in R, playing around with the max lag length. As I was changing the length, I got mixed results. Below is the output for lag.length=16 (producing normal results), lag.length=20 (warning message: "NaNs produced") and lag.length=25 (also producing normal results).

I would greatly appreciate if somebody could explain what could be the reason of NaNs being produced on lag.length=20 but no such thing happening on 16 (smaller lag length than 20) and 25 (larger length than 20). If I am missing anything, other suggestions are also appreciated.

R output below:

---

#max.lag=16
varBTC <- VAR_dataBTC_ts %>%
VAR(lag.max=16,ic="AIC")

#max.lag=20
varBTC <- VAR_dataBTC_ts %>%
VAR(lag.max=20,ic="AIC")
Warning messages:
1: In log(sigma.det) : NaNs produced
2: In log(sigma.det) : NaNs produced
3: In log(sigma.det) : NaNs produced

#max.lag=25
varBTC <- VAR_dataBTC_ts %>%
VAR(lag.max=25,ic="AIC")

---