Glass Comparison: RI, Density and LA-ICP-MS

Before any instrumental measurement, the analyst conducts a macroscopic and low-power microscopic examination of the questioned (Q) and known (K) glass fragment populations. This examination addresses properties visible without chemical testing: colour, texture, surface finish, thickness, curvature, fluorescence, and the presence of coatings or interlayer residue.

Colour in glass comes from two sources: transition-metal or rare-earth colorant additions (cobalt for blue, chromium for green, iron for green-amber, manganese for purple) and the intrinsic colour of the base glass (soda-lime float glass has a slight green tint from residual iron, visible in thick sections or when viewed edge-on). A colourless fragment cannot be consistent with a green bottle, and a deep blue fragment cannot be consistent with a standard window pane. This basic exclusion is made in seconds under white-light illumination.

Fluorescence examination under 254 nm UV illumination identifies tin-side surfaces in float glass (blue-white fluorescence, as described in the companion glass-types topic), borosilicate glass (often weakly fluorescent), and coated glass (various coatings fluoresce depending on their composition). The tin-side examination is particularly useful when one fragment is oriented (face known) and the other must be oriented to match: if both fragments' fluorescent faces correspond, their spatial relationship in the original pane is consistent.

Thickness measurement with a micrometer gives a population distribution for each fragment set. Float glass is manufactured in defined nominal thicknesses (2 mm, 3 mm, 4 mm, 5 mm, 6 mm, 8 mm, 10 mm, 12 mm), and a questioned fragment cluster centred at 4.1 mm cannot be from a 6 mm window. Within the same nominal thickness class, fragments from the same pane will cluster narrowly around the same value, while fragments from different panes of nominally the same thickness will show slightly different mean values. The overlap depends on manufacturing tolerances; the FBI glass protocol specifies that thickness measurements within 0.1 mm of each other are treated as consistent.

The overall physical feature examination is documented in the examination record before any RI or elemental measurement is made, preserving the integrity of the sequential gate design. An exclusion at the physical level terminates the examination; a pass at the physical level proceeds to RI.

The Glass Refractive Index Measurement instrument (GRIM-3, manufactured by Foster + Freeman Ltd in the UK) automates the phase-contrast-microscopy approach to refractive index measurement. A single glass fragment (minimum approximately 0.5 mm) is placed on a heated stage in contact with an immersion oil of known RI. The stage temperature is varied while the microscope monitors the phase-contrast appearance of the fragment-oil interface. When the glass RI matches the oil RI (at the current temperature), the fragment becomes invisible in the field of view (the "match temperature"). The oil RI as a function of temperature is known from calibration, so the match temperature gives the glass RI directly.

The ASTM E1967 standard (Standard Test Method for the Measurement of the Refractive Index of Glass Specimens by Automated Phase-Contrast Microscopy) specifies the calibration requirements, measurement replication, and reporting format for GRIM-based RI determination. The standard requires a minimum of three measurements per fragment (or per fragment population if fragments are too small for individual measurement) and specifies the immersion oil certification requirements. The stated within-laboratory reproducibility of the GRIM-3 at standard conditions is approximately 0.0001 RI units (one unit in the fourth decimal place at 589 nm), which is the measurement uncertainty that determines the comparison criterion.

The match criterion most widely used in the forensic glass community is based on the within-population RI variation of the questioned and known fragment populations. The ENFSI ENG3-2013 guideline specifies a resampling-based comparison (the within-population distributions of Q and K are compared using a two-sample test, with the critical value set at a significance level of typically 0.05 or 0.01). In the US, the FBI and SWGMAT guidelines use a conservative criterion based on ±2 standard deviations of the within-population distribution: if the Q and K population means are within 2 standard deviations of each other (accounting for both populations' variances), the populations are treated as analytically consistent.

An important limitation of RI comparison is that the soda-lime glass manufactured in a particular geographic region or time period clusters within a narrow RI range, meaning that a large proportion of randomly selected soda-lime glass fragments from different panes will have overlapping RI distributions. Studies of glass RI distributions in New Zealand and the UK (Buckleton, Triggs, and Walsh 1994, 1995) and the UK (Lambert and Evett 1984, Curran et al. 2000) have found that the RI distribution of common window glass in a given population is unimodal and relatively narrow, which reduces the evidential significance of an RI match when considered alone. This is why LA-ICP-MS elemental analysis was developed: to provide discrimination within the RI-indistinguishable population.

Density measurement provides a second independent physical property for glass comparison. Module 4 of this subject covers the sink-float and density-gradient column methods in detail. In the glass comparison workflow, the density-gradient column is the standard method: a linear density gradient is prepared in a glass tube using a mixtures of bromoform (density 2.89 g/cm3) and bromobenzene (density 1.50 g/cm3) or similar heavy-light pairs, producing a stable column in which the density increases continuously from top to bottom.

When a glass fragment is placed in the column, it sinks until it reaches the depth at which the column density equals the fragment density, then remains in suspension at neutral buoyancy. The depth of neutral buoyancy is read against a calibration scale (established by floating density reference pellets of known density). The precision of density-gradient column measurement is approximately 0.0001-0.0002 g/cm3 under well-controlled conditions, which is comparable to the precision available from oscillating-U-tube digital densitometers for larger samples.

Density adds discrimination beyond RI because the two properties are not perfectly correlated within a glass composition: different manufacturing batches of soda-lime glass at similar RI can have slightly different density values, and vice versa. A two-dimensional RI-density comparison space therefore has more discriminating power than either property alone. Research from the Forensic Science Service (Buckleton, Walsh, and colleagues) established the statistical framework for two-dimensional RI-density comparison in the 1990s.

The practical limitation of density measurement is the sample size requirement. The density-gradient column method requires fragments large enough to be placed individually in the column and observed at neutral buoyancy: fragments smaller than approximately 1 mm are difficult to handle reliably. In casework involving fine glass fragments (such as those transferred to clothing in secondary contact), the fragment population may be too small for density measurement, leaving RI as the only physical-property method and directing the analysis to LA-ICP-MS for discrimination.

Laser Ablation Inductively Coupled Plasma Mass Spectrometry (LA-ICP-MS) combines three instrumental components. The laser ablation (LA) front end uses a focused pulsed laser beam (typically a frequency-quadrupled Nd:YAG at 266 nm or an ArF excimer at 193 nm) to ablate a microscopic crater (typically 50-100 micrometres in diameter, 5-20 micrometres deep) in the glass sample surface. The ablated material is transported as an aerosol to the ICP torch. The inductively coupled plasma (ICP) atomises and ionises the aerosol at approximately 7000-8000 degrees Celsius. The mass spectrometer (MS) separates the resulting ions by mass-to-charge ratio and measures their signal intensities, which are proportional to element concentrations.

The ASTM E2927 standard (Standard Test Method for Determination of Trace Elements in Glass Samples Using Laser Ablation Inductively Coupled Plasma Mass Spectrometry) specifies the minimum requirements for forensic glass comparison by LA-ICP-MS: the element suite (28 elements when the full extended REE complement is run, namely Li, Na, Mg, Al, K, Ca, Ti, Mn, Fe, Rb, Sr, Y, Zr, Ba, La, Ce, Pr, Nd, Sm, Eu, Gd, Dy, Er, Yb, Pb, Bi, Th, U; the core ASTM E2927 mandatory subset is 18 elements), calibration approach (typically using NIST SRM 612 glass standard as the primary calibration material and NIST SRM 614 or 1831 as verification standards), number of replicate ablations per sample (minimum 3, typically 5-7), and the data-reduction procedure.

The choice of elements includes both the major/minor elements that carry bulk-composition information (Na, Ca, Mg, Al, K, Fe, Ti) and the trace elements whose concentrations are most variable between manufacturing sources (rare-earth elements La, Ce, Nd, Sm; heavy metals Pb, Bi; and transition metals Rb, Sr, Zr, Ba, Mn). The rare-earth elements are particularly discriminating because their relative abundances reflect the geochemical source of the silica sand raw material, which varies between quarry locations. Two float-glass plants using sand from different quarries will have systematically different rare-earth element ratios even if their major-oxide compositions are similar.

The ASTM E2927 method was validated in an interlaboratory study involving the FBI Laboratory, the ATF Laboratory, the Florida International University Trace Evidence Analysis Facility, the RCMP National Forensic Laboratory, and several university research groups. The results, published by Trejos, Almirall and colleagues in the Journal of Analytical Atomic Spectrometry (2013), established the within-laboratory and between-laboratory precision of the method and the discrimination power available from the multi-element suite.

An LA-ICP-MS run ablates approximately 10 nanograms of material from a 100-micrometre crater. This is essentially non-destructive: the crater is invisible to the naked eye and the remaining fragment is preserved for additional testing or reanalysis. One fragment can sustain multiple ablations (at different locations to assess within-fragment homogeneity) and still be available for archiving or physical examination.

The elemental data from an LA-ICP-MS comparison consist of vectors of 27 or more element concentrations for each glass fragment population. Comparing two populations of such vectors requires a multivariate statistical test, not a series of independent univariate comparisons. The key reasons are: (1) the elements are correlated (a glass with high Ba tends also to have high Sr, because both are incorporated together in the feldspar component of the raw sand), so treating them as independent tests inflates the apparent discriminating power; (2) multiple independent tests at alpha = 0.05 accumulate false-positive rates rapidly (with 27 tests, the overall false-positive rate is approximately 75% if the tests were truly independent and alpha = 0.05 each).

Hotelling's T2 test is the multivariate generalisation of the Student's t-test. It tests whether the mean element-concentration vector of the questioned fragment population is consistent with the mean vector of the known fragment population, accounting for the full covariance structure of the element measurements. Under ASTM E2927, the Hotelling T2 comparison at a specified significance level (typically alpha = 0.05) is the recommended match criterion for the major-element subset. If the T2 statistic exceeds the critical value, the populations are declared analytically distinguishable.

The Mahalanobis distance is a related measure that expresses the distance between two multivariate observations in units of the population's standard deviation, accounting for covariance. A Mahalanobis distance of 1.0 means the two observations are one (multivariate) standard deviation apart. The Mahalanobis distance is used as a dissimilarity measure in nearest-neighbour and classification approaches to glass source attribution: questioned fragment vectors are compared against a database of known glass populations, and the most similar known population (smallest Mahalanobis distance) provides the best-match candidate.

The FBI Laboratory glass comparison protocol published by Koons, Buscaglia, Bottrell, and Miller (in Forensic Science Handbook, 2nd ed., ed. Saferstein, Prentice-Hall, 2002, vol. 1, pp. 161-213) specifies a conservative range overlap criterion for the major-element comparison and a Hotelling T2 comparison for the full multi-element suite. The ENFSI ENG3-2013 guideline recommends the multivariate statistical framework without specifying a single preferred test, acknowledging that the field is still developing consensus on the best approach.

Principal Components Analysis (PCA) is frequently used as a visualisation and exploratory tool alongside the formal hypothesis test. A PCA biplot of the first two or three principal components of the element dataset allows the analyst to visually assess whether Q and K fragment populations cluster together or separate, before applying the formal statistical test. PCA does not replace the hypothesis test but provides an accessible visual representation for reports and courtroom presentations.

The standard casework workflow for glass comparison follows a sequential gate design prescribed by ENFSI ENG3-2013, ASTM E2927, and the SWGMAT guidelines. The gates are applied in the order: physical features, then RI (if physical features are consistent), then density (if RI is consistent and sample size allows), then LA-ICP-MS (if density is consistent or sample size precludes density, and RI alone has insufficient discrimination power).

The first gate (physical features) eliminates obvious mismatches: colour class, thickness class, surface type. A fragment that is visually orange cannot be from a clear window pane; a fragment that is 10 mm thick cannot be from a 4 mm window. These eliminations require no instrumentation and take minutes.

The second gate (RI by GRIM-3) is applied to all fragments that pass the physical gate. Each fragment is measured to ASTM E1967 specifications. The Q and K population distributions are compared. If the populations are analytically distinguishable, the examination terminates with an exclusion. If the populations are analytically indistinguishable, the examination proceeds.

The third gate (density by density-gradient column) is applied when fragments are large enough and when additional discrimination is needed. A Q/K population that passes the RI gate but diverges in density is excluded. A population that passes both RI and density gates proceeds to elemental analysis.

The fourth gate (LA-ICP-MS) is applied to all submissions where the RI and density gates are passed and a high-discrimination comparison is needed (or where sample size precludes density measurement). The ASTM E2927 element suite is measured, the Hotelling T2 comparison is performed, and the result is reported as analytically indistinguishable or analytically distinguishable at the stated significance level.

Exclusion at any gate is a final determination: the analytical data are inconsistent with a common origin. Indistinguishability at all gates is not proof of a common origin; it is a statement that the analytical methods available cannot differentiate the two fragment populations, and the significance of that statement depends on the Bayesian likelihood ratio.

The Bayesian likelihood ratio (LR) provides the formal bridge between an analytical result and a forensic opinion. The LR is defined as the probability of the observed evidence (the glass fragment measurements) given the prosecution hypothesis (Hp: the questioned fragment came from the same source as the known fragment) divided by the probability of the observed evidence given the defence hypothesis (Hd: the questioned fragment came from some other, unrelated source).

LR = P(E | Hp) / P(E | Hd)

A LR of 1 means the evidence is equally probable under both hypotheses and is therefore neutral. A LR of 100 means the evidence is 100 times more probable if the association is true than if it is not, which is a moderate level of support for the prosecution hypothesis. A LR greater than 10,000 is strong support; the ENFSI ENG3-2013 guideline provides a verbal scale for translating LR values into reported opinion strength (limited, moderate, moderately strong, strong, very strong support).

For glass RI comparison, the numerator of the LR (P(E | Hp)) is estimated from the within-source variation of the known glass: how often do fragments from this source show RI values as similar as the Q and K populations? The denominator (P(E | Hd)) is estimated from the distribution of RI values in the relevant glass population in the environment: how often does a random glass fragment from another source have an RI in the observed range? This denominator requires either a database of glass RI values representative of the local environment or a theoretical model of the RI distribution. Databases have been developed for the UK (Lambert and Evett 1984, updated by Curran 1997 and Aitken, Curran, and Buckleton 2004), the US (Buscaglia 1994, Almirall 2003), and Europe generally (ENFSI collaborative studies).

For LA-ICP-MS, the LR is computed in multivariate element space using the same logic but with 27-dimensional covariance matrices. Software implementations including the ENFSI-supported LR calculator and the glass-comparison software developed at Florida International University implement the multivariate LR computation from ASTM E2927 output data.

The FBI Laboratory, following the OSAC-endorsed reporting guidelines, uses a conservative categorical scale rather than a numerical LR for public-facing glass-comparison reports, similar to the approach used in DNA mixture reporting: "The glass is consistent with / could have come from / is not associated with the known source." The more quantitative LR approach is embedded in the supporting technical documentation and available for expert testimony. The UK Forensic Science Regulator's 2020 and 2022 guidance on LR reporting in forensic science strongly favours the numerical LR approach for disciplines with validated probabilistic frameworks, and glass comparison is one of the disciplines where that validation exists.

In jurisdictions where quantitative LR reporting is not yet routine, analysts express the result qualitatively as "consistent with" or "inconsistent with" a common source and may introduce the numerical LR framework where the strength of trace evidence is specifically challenged.

In Canada, RCMP glass comparison reports use a verbal scale aligned with ENFSI guidelines, with the numerical LR available for court testimony. The RCMP National Forensic Laboratory published research (Corcoran, Dewsbury, and Curry) on multivariate glass comparison that informs the Canadian operational standard.

A complete glass comparison report addresses four elements: the analytical findings (specific RI values with measurement uncertainty, density values, element concentrations), the comparison result (indistinguishable or distinguishable, with the test applied and significance level), the probabilistic interpretation (likelihood ratio or verbal equivalent, with database source), and the limitations (sample size, secondary transfer probability, background glass-fragment level).

The secondary transfer limitation must be addressed explicitly in any opinion about association. As established in UK Forensic Science Regulator guidance, a glass match between a questioned fragment and a known source does not establish direct contact between the suspect and the source pane: the fragment may have transferred to a secondary surface and then to the suspect. The report should state whether secondary transfer is a realistic alternative explanation given the case circumstances and, if possible, reference published data on secondary transfer rates.

Background glass-fragment levels are relevant because they affect the denominator of the LR. A suspect who works in a glazing workshop will have higher background glass-fragment levels on their clothing than a suspect who works in an office. The significance of finding glass that matches a burgled window is considerably lower if the suspect's occupation involves regular glass handling. This should be addressed in the report with reference to any background-level measurements made from control samples.

The cross-examination challenge most frequently mounted against RI comparison is that a large proportion of common window glass is analytically indistinguishable by RI: the LR denominator is high because many unrelated panes have similar RI values. This argument is correct for RI alone, which is why LA-ICP-MS is applied when discrimination power is needed. The cross-examination challenge against LA-ICP-MS is that the test is highly discriminating and therefore easily prejudicial: if the jury perceives the method as infallible, they may over-weight the evidence. The expert witness's role is to convey the actual LR value and the assumptions behind it, so that the court can correctly calibrate the evidence's weight.

The ENFSI ENG3-2013 guideline, the ASTM E2927 standard, and the OSAC Glass Subcommittee guidelines in the US all require the analyst to state the limitations of each measurement tier, the sample size and statistical power of the comparison, and the assumptions underlying the LR calculation. Reports that assert conclusions without stating assumptions are vulnerable to Daubert or Frye challenge in the US and to exclusion under the Criminal Procedure Rules Part 19 in England and Wales.