Transitions between adjacent cells with finite and infinite values are missing.

[famuvie]

Hi! First, thanks a lot for the work put into the fantastic package {gdistance}!

I wanted to report a behaviour that I think should be considered a bug, concerning transitionLayer.

Given two adjacent cells with finite values $x$ and $y$, the transition between them is $f(x, y)$, where $f$ is the transition function. Say that $f$ is the mean() function, for instance. When $y$ grows bigger, the transition grows bigger as well. So, if $y = \infty$, I would expect the transition to be $\infty$, for consistency.
Yet, the result is a missing value, as if the two cells were not connected, or as if $y$ was NA.

A MRE follows:

## MRE gdistance::transition
library(gdistance)
#> Loading required package: raster
#> Loading required package: sp
#> Loading required package: igraph
#> 
#> Attaching package: 'igraph'
#> The following object is masked from 'package:raster':
#> 
#>     union
#> The following objects are masked from 'package:stats':
#> 
#>     decompose, spectrum
#> The following object is masked from 'package:base':
#> 
#>     union
#> Loading required package: Matrix
#> 
#> Attaching package: 'gdistance'
#> The following object is masked from 'package:igraph':
#> 
#>     normalize
m0 <- matrix(1:4, 2, 2)
r0 <- raster(m0)

m2 <- m1 <- m0

m1[2, 2] <- NA
r1 <- raster(m1)

m2[2, 2] <- Inf
r2 <- raster(m2)

tr0 <- transition(r0, mean, directions = 4)
tr1 <- transition(r1, mean, directions = 4)
tr2 <- transition(r2, mean, directions = 4)

transitionMatrix(tr0)
#> 4 x 4 sparse Matrix of class "dsCMatrix"
#>                     
#> [1,] .   2.0 1.5 .  
#> [2,] 2.0 .   .   3.5
#> [3,] 1.5 .   .   3.0
#> [4,] .   3.5 3.0 .
transitionMatrix(tr1)
#> 4 x 4 sparse Matrix of class "dsCMatrix"
#>                 
#> [1,] .   2 1.5 .
#> [2,] 2.0 . .   .
#> [3,] 1.5 . .   .
#> [4,] .   . .   .
transitionMatrix(tr2)
#> 4 x 4 sparse Matrix of class "dsCMatrix"
#>                 
#> [1,] .   2 1.5 .
#> [2,] 2.0 . .   .
#> [3,] 1.5 . .   .
#> [4,] .   . .   .


## Inputs are different, but results are the identical
identical(r1, r2)  # FALSE
#> [1] FALSE
identical(transitionMatrix(tr1), transitionMatrix(tr2))  # TRUE
#> [1] TRUE

sessionInfo()
#> R version 4.5.1 (2025-06-13)
#> Platform: x86_64-pc-linux-gnu
#> Running under: Linux Mint 22.1
#> 
#> Matrix products: default
#> BLAS:   /usr/lib/x86_64-linux-gnu/blas/libblas.so.3.12.0 
#> LAPACK: /usr/lib/x86_64-linux-gnu/lapack/liblapack.so.3.12.0  LAPACK version 3.12.0
#> 
#> locale:
#>  [1] LC_CTYPE=en_GB.UTF-8       LC_NUMERIC=C              
#>  [3] LC_TIME=en_GB.UTF-8        LC_COLLATE=en_GB.UTF-8    
#>  [5] LC_MONETARY=fr_FR.UTF-8    LC_MESSAGES=en_GB.UTF-8   
#>  [7] LC_PAPER=fr_FR.UTF-8       LC_NAME=C                 
#>  [9] LC_ADDRESS=C               LC_TELEPHONE=C            
#> [11] LC_MEASUREMENT=fr_FR.UTF-8 LC_IDENTIFICATION=C       
#> 
#> time zone: Europe/Paris
#> tzcode source: system (glibc)
#> 
#> attached base packages:
#> [1] stats     graphics  grDevices utils     datasets  methods   base     
#> 
#> other attached packages:
#> [1] gdistance_1.6.4 Matrix_1.7-4    igraph_2.1.4    raster_3.6-32  
#> [5] sp_2.2-0       
#> 
#> loaded via a namespace (and not attached):
#>  [1] digest_0.6.37     codetools_0.2-20  fastmap_1.2.0     xfun_0.52        
#>  [5] lattice_0.22-7    magrittr_2.0.3    glue_1.8.0        knitr_1.50       
#>  [9] pkgconfig_2.0.3   htmltools_0.5.8.1 rmarkdown_2.29    lifecycle_1.0.4  
#> [13] cli_3.6.5         terra_1.8-60      grid_4.5.1        reprex_2.1.1     
#> [17] withr_3.0.2       compiler_4.5.1    tools_4.5.1       evaluate_1.0.5   
#> [21] Rcpp_1.1.0        yaml_2.3.10       rlang_1.1.6       fs_1.6.6

^{Created on 2025-09-02 with reprex v2.1.1}

[andrewmarx]

Is there a practical reason why you would want infinite values in the transition matrix? My initial thought is that having relatively large values will be a sufficient alternative.

Off the top of my head, infinite values will definitely cause issues with row-normalization. I also suspect it might cause issues with some of the linear algebra stuff, particularly anything involving decomposition.

It might be better if I change it so that, rather than replacing the infinite values with 0, the transition() function simply fails. But I would need to spend some time looking into that first to be sure there aren't any unintended consequences.

[famuvie]

No, I don't have any practical application of using infinite values in a raster (and in a transition matrix). It is more a matter of consistency and expected behaviour.
I reviewed some code that involved a cost surface with values of 0 instead of NAs in large regions, and I raised a warning because I expected the cells adjacent to such areas to behave as if they were the same point (infinitely close points, anyway), while in reality they should be very far apart. So I thought the results were rubbish, but we found out that once fixed, the results were exactly the same.

I just dug until I found where the problem was. It turned out not to be a problem. Still, it does not seem sensible for two such different situations to lead to the same results.

If handling Infinite values causes numerical problems, I find it more reasonable to just exclude such situation from the domain of the function and fail with an informative message, rather than changing the values to something else without notification. But I understand the compromise.

I appreciate you having looked at it. Thank you for your answer.