Pitfalls Checklist

The tables below organize input and post-processing errors that frequently occur in transport calculations, indexed by symptom. When a calculation looks wrong, check against these tables before reaching for a physical interpretation.

TranSIESTA / TBtrans

Symptom	Cause	Prescription	Reference
$T(E)$ unchanged even when TBT.kgrid_Monkhorst_Pack is changed	The block was placed in TBtrans.fdf — TBtrans automatically reduces the transport-direction k to 1	Place the block at the very end of input.fdf	07
Finite bias applied, but the result is identical to 0 V	TS.Voltage was added only in input.fdf — among duplicate fdf keywords the definition read first prevails, so the 0 eV in TS.fdf remains in effect	Edit both TS.Voltage in TS.fdf and TBT.Voltage in TBtrans.fdf, then verify with grep -i voltage *.fdf	09
You have to calculate the 0 V and re-use the TSDE error right after a bias calculation starts	trans.TSDE from the previous bias was not copied	Copy trans.TSDE (mandatory), trans.DM, and trans.XV from the immediately preceding bias	09
Bias calculation crashes immediately with a header mismatch	Bias-dependent caches such as trans.TSGF* and trans.TSCC* were also copied	Never copy the caches — they are regenerated automatically	09
trans.TBT.nc is several GB, TBtrans OOM	Heavy output options such as TBT.T.Bulk, TBT.T.Eig, TBT.Current.Orb	Keep the minimal option set — sisl post-processing is sufficient for eigenchannels	07, 10
$T(E)$ that should be ballistic decays exponentially with device length	Excessive $\eta$ — artificial decay caused by over-spread sharp resonances	Lower $\eta$ and check convergence, shrinking the energy grid delta along with it	η broadening
Eigenchannel interpretation is ambiguous	The chosen energy, bands, PDOS, and channel symmetry were not examined together	State the energy appropriate to the analysis goal and verify $\sum_i\tau_i=T$	10
A symmetric junction shows rectifying I–V	Asymmetric window in the zero-bias integral ($\mu_L=0$, $\mu_R=-V$)	Integrate with the symmetric window $\mu_{L,R}=\pm V/2$	08

Common to SIESTA / VASP

Symptom	Cause	Prescription	Reference
Energy differences between structures are at an irreproducible noise level	SCF.DM.Tolerance loosened (e.g., 1.0d-4)	Keep the tutorial standard 1.0d-8 — it is the premise of energy comparison	01
Pseudopotential file not found abort right after the start	Mismatch between the species in ChemicalSpeciesLabel and the .psml files — especially after transferring to another machine	Make a habit of a 1:1 ls check of species ↔ pseudo files before submission	Setup
(VASP) XC contribution missing from the LOCPOT used for vacuum alignment	LVHAR=T overrides LVTOT=T — turning both on saves only Hartree+ionic	Use only the one flag that matches the purpose: LVTOT for the total potential, LVHAR for Hartree only	03
Several calculations submitted with a shell loop, but only the first case runs	ssh/mpirun inside the loop consumes the loop's stdin	Use ssh -n or redirect < /dev/null	12

Common principles

• Confirming existence is not confirming operation — a flag being present in the input does not mean the feature is on. Grep the stdout echo to confirm it was actually applied.
• Record the provenance of inputs — leaving even one line on which script generated them and with what physical intent (substitution site, vacuum, lattice reference) lets you immediately decide "input bug or physics?" when a result looks odd.
• Inspect templates for outliers when copying — when reusing someone else's input, check the key five lines to see whether the convergence criteria, lattice, and output options match your own purpose.